Global AI Bootcamp 2025

All right, everybody, we're gonna get started here in just a second. I'm gonna stop sharing my screen and let you see my face. Hopefully. Oh, there we go. Hi, everybody. Great to have you here. Welcome to the 2025 Global AI Bootcamp. It's gonna be an amazing day. We, what I'm gonna do right now is actually just send out the link to make sure that, oh, we got a pretty good representation here. So that's amazing. So I'm gonna send out the link. We're gonna get started here in just a second. I'm gonna stop sharing. I'm still— sorry, I was listening to the LinkedIn, and I just went live on LinkedIn. Okay, cool. So I'm gonna actually, for just a moment, if you will allow me to make sure everyone has the right link to join. So I'm gonna post this in the school community, and then also I will be posting it online just to make sure that nobody— we had about 150 people sign up. And so I wanna make sure nobody's like, where is it? I can't connect. I can't find the link. You know how that is sometimes. So I'm gonna pull that up right now, and I had to stop sharing in order to, to see it and make sure that you all— oh, I'm in the Global AI Bootcamp. Oh, there we go. Make sure everyone has access, and I didn't necessarily want you to see everyone's emails. There we go. Let's see here. Organizers information. Oh no, no, no. Let's try this. Nope.

All right.

Global AI Community. I'm just gonna do— there we go. Community. Um, just have to log in real quick. Thank you for your patience. And then we're gonna get started with I just wanna make sure everybody can see it. Let me go to my profile. Amazing. Alright, I'm gonna— oh, I just downloaded them again. We had even more registrations show up last night. Let me go ahead. And so if you get this email, I apologize for inundating you with all the info. I just wanna make sure everyone has a chance to be here who wanted to be. And oops, Ctrl+Z. So I'm gonna send a quick email out to my friends to make sure the 100+ of you who signed up got a chance to join us. This is gonna be recorded and chopped up and made a little bit more production savvy by the end. So you'll have a chance to watch these on the replay. It is a pretty long day. We have 4 sessions, a couple of guest speakers, as well as some labs. But y'all know me, I like to do live labs. So we will also be doing these labs in the AI Leadership Institute for all of the members. We'll be going through these labs step by step. Here, I'll just be giving the labs to you, and it's kind of do-it-yourself style. Okay. Someone's calling me from here. I wonder if they're here. Oh, let's see. I'm gonna answer it. Hello? Oh, hi. Sorry, I can't talk right now. Thank you. Bye. Um, sorry, I'm on stage. I thought it might be somebody, but no, it was a salesperson. No, not for you. Not right now. Um, let's see here. Oh my gosh, so many emails, so little time.

There we go.

All right, so let me BCC y'all. There we go. Don't look.

There we go.

Starting now. A global AI boot camp. All right, so I'm just going to throw in the invite from our Zoom. Oh, good. More people have joined us. Welcome, everybody. Nice to see you all. Copy the link. I do want to make sure that if you are online and watching this, that you have registered for this event, and that's the only way you're gonna get access to the recordings as well as to the lab documents. So I'm super excited about that. All right, join us now. All right, so only the people that registered do get access to all of this. You also will get the coveted badge. So you— there is a Global AI Bootcamp badge that is only available to people who have gone through this experience, and you will be able to say. I'm gonna encourage you to say that you did this on LinkedIn as well. But we are live. Welcome, welcome. I'm super excited. So last year I did this, and I only did it online, and I only did it to— I only did it with people that I knew. I didn't open it up to like the world. So it's pretty cool because there's— oh my gosh, my hair. There's a ton of people that are here that, you know, you found it organically. So that's kind of fun. Speaking of which, don't forget Adam's in the house. What's up, Adam? And Michelle, Mauricio's here. A bunch of Miami people. If you want to come down to the AI Center, you're welcome to show. Like, come on, come on. It's not bad. It's not terrible. Oopsie. Is that you? Did you join the Zoom?

Yeah.

Yes.

Oh, sorry.

Okay. All right, cool. Oh, we're up. Oh, oh, I just resent the link and more people are joining us. So good, it worked. It worked. Okay. Oh yeah, you're muted. So this is a pretty cool environment. It's not a webinar as I typically do. It's actually just a good old-fashioned call. So you— sorry, but I will have to mute y'all if you get loud and rowdy. So I'm gonna mute everybody now because usually you don't mean to be unmuted. But that is the danger of doing calls instead of webinars is that people accidentally unmute themselves or intentionally, and it kind of messes up the, like, flow for everyone else. And so we're gonna try and just make sure we will have time for questions at the end of every session. There'll also be an opportunity. I actually got for everyone who is currently on the line, I have credits. That Azure has— Microsoft Azure team has given to us. So that's exciting. So when we get to the end where we're doing labs, we have 2 different lab workshops that we're going to do. One is our classic AI Bot Builder. For those of you who are not familiar with the 5 Ps of prompt engineering and the AI safety system, we're gonna go through all of those things that I'm deeply passionate about and that are in my new book that just came out, which of course, But that's okay. But it comes out tomorrow. Tomorrow is the big day that that book launches, even though I've gotten some early copies and been able to take them to a couple of recent events. But that being said, we're gonna have that piece, but we also have some things that only our AI Leadership Institute members have seen. So how do you build an AI— build and deploy an AI model with no code whatsoever? So that's inside of AI builder tool. And then the other part is how to actually build an— or we'll call it vibe coding. So I'm gonna give you an introduction to this concept of vibe coding, which means you vibing with the machine to build software. And it's very much for those of us who are software engineers or want to be, but wanna take advantage of the ability to have machines write the code for you while you do the ideation and creation process. It's very cool, very fun. I've done a ton of it. And it's really good for businesses that you're, you know, wanna build assets that they wanna deploy on a website that used to take weeks, months, or years that now we can do in days, maybe even hours. So really exciting. Alright. So welcome everyone to the Global AI Bootcamp. We are going to kick off this amazing session with the required keynote. So there is a recorded keynote that I'm gonna share with you. Everyone who has attended the Global AI Bootcamp worldwide watches this keynote. So you all have the same kind of context. I do know everyone who's been— who is here. I have either worked with directly, communicated to directly, been in a room with, meetings with, been on their podcast. So I am very big fan of everybody who will be speaking in the keynote and really looking forward to it. So I'm going to stop sharing my screen and share the keynote. And then after the keynote, we'll begin our direct sessions, which will be me kind of whiteboarding for you. And, and maybe we'll do a couple presentations. They did so much work to put this together. I do wanna give huge props and shout out over to the team at the Global AI Community and invite you all. I mean, I created, I run now the Miami Global AI Community. But you can build your own chapter in whatever city you live in. So just know that that is a place you can go. There's huge, So there's lots of different locations all over the world. But this is something every year that we do together. We all teach the same stuff, and we bring our own communities together. So if you want to start a chapter and you want my help, of course, please reach out. I'd be happy to help you do that. Alright, good to see some friends in the house.

What?

Eric's in the house. Good to see you, Eric. It's been a while. Awesome. There's Alonso, and then Julia's here. Mauricio, I mentioned. Michelle, good to see you. Rob, what's up?

Salim!

Oh my gosh, Salim is here. See, because we did something in a good time frame for you. So that's awesome. Nice to see you. Sky is in the house. Hello, Tony. Yeah, what's up? Oh my gosh, it's so good to see you all. So, and we have more people coming because I just sent the list out to everyone. So they'll be joining us. But let's go ahead and kickstart this fun, fine day with the keynote. Like I said, it is required viewing. It's not too long, but you know, I could play at 2x speed. That's what I do. But I'm gonna encourage you all to watch it. But it's fun. Let's take a look. I'm gonna go ahead and share my screen, make sure I share the appropriate audio. If I can find it. I don't remember where I put it.

Okay, hold on.

Oh, that's right. I shifted screens.

Here we go.

I think it's this one. I'm not 100% sure. All right, we're gonna try it. Let me go down and make sure. Okay, I'm gonna try this. Oh, darn, can't do that. See, I had it going really well first. I'm gonna try one more time.

Share.

And I think it's— oh no, now there are two that are black. Okay, hold on, let me go back. Oh, there it is. I found it. Alright. And then share audio. We should be good. There we go. Alright. So this, I'll let her introduce herself, but she is new to Microsoft and she is a really cool leader. She is actually a peer to Eric Boyd. Eric Boyd runs engineering. That's the team that I worked on. Asha runs now product. So she's going to introduce the entire day to you. So without further ado, I am gonna need you all to confirm in the chat that you can see this and hear it. So please, when you can hear it, go ahead and throw it into the chat. I hear you, or this is awesome, or whatever you wanna do to let me know that you can hear what's going on.

Let's go.

Hi, I'm Asha Sharma, Corporate Vice President of AI Platform at Microsoft. And every day I get to work with teams that are pushing the boundary of what AI can do. We're helping developers, scientists, businesses, and people around the entire world use AI to unlock new possibilities. AI is changing the way we build. It's changing the way we create. And how we innovate. And as developers, you are at the heart of this transformation, whether you're working on applications, optimizing workflows, or pushing the boundaries of what is possible. AI is no longer a future technology. It is here. It is right now. And it is shaping the way that we work and the way that we live. Now, the beauty of AI is that it's not just for researchers and scientists in a lab. It's actually for every single developer. From integrating AI-powered tools in your daily workflows to building the next generation of applications, AI is an amplifier of human potential. And the best part? The tools and models are available today to make it more accessible than ever before. So in this keynote, we're going to take you on a journey through AI for developers. You're going to hear from some the brightest minds in the industry about why AI matters, how Python has been so foundational to AI's rise, the latest advancements in AI agents, and how we can all build AI responsibly. The opportunity is in front of us and it's massive. As developers, you're not just coding, you're shaping the future. Whether you're just getting started with AI or you're already building with it, there has never been a better time to dive in. So welcome to the Global AI Bootcamp 2025.

Thank you, Ayesha, for setting the stage for today's keynote. AI is transforming the way we build software, and as developers, we have a huge opportunity to be part of this revolution.

I'm Roland, and this is Sami, and we'll be your hosts as we travel to speak with some of the most influential voices in AI, exploring how it's shaping development what's next, and how we can all get started.

Behind the scenes, our third board member and producer making all of this possible, Henk, who's ensuring we capture every moment and bringing you the best conversations.

This keynote is part of the Global AI Community, an independent organization with over 85,000 members worldwide. Right now, all over the world, bootcamps are taking place in over 100 locations just like this one. And together we're reaching more than a quarter million people around the world.

Now, this is more than just an event. It's a movement, a global effort to help developers of all backgrounds to learn, build, and grow in AI. And today we're kicking things off with someone who knows exactly why developers need to embrace AI, Scott Hanselman.

Scott is a developer, educator, tech leader who's passionate about making AI accessible to everyone. So let's dive into why learning AI is essential for developers. Hi Scott, we're here with the global AI community at the Global AI Bootcamp. Could you please help us make sense of what's happening in the AI space?

I think it's getting a little out of control. I know that everyone's excited about AI, but I think it's a great opportunity at the AI Bootcamp for everyone to just chill out a little bit. You don't have to put AI on everything. Additionally, I think it's important for us to remember that AI sits on top of data science, on machine learning, on deep learning, on statistics itself. And while it feels like AI just came out last year and we should all pivot and run in the direction of AI. In fact, it's been around for a very long time and there's a lot of really great experts, some of which we've collected today, that can put this stuff into context. So I'm excited about AI, but let's all just chill.

And during this keynote, we're coming from several different angles, and hopefully you can shed some light on the developer side of things. Why should they start learning AI today?

Well, I remember when I went from a black and white screen to a color screen and I got syntax highlighting and I was looking at code that wasn't all one color. And then I got autocomplete and then I got IntelliSense. And now I've got spicy autocomplete because I've got GitHub Copilot that can help me as a developer, not just generate code, but better understand my code. I tend to use the Socratic method and I interview GitHub Copilot as a— as if it were an intern that was helping me with my code. Developers should think about what AI could what it can do for them while they're writing code, what it can do for their DevOps experience while they're maintaining their sites, and then what it can do for their customers. Because not everything needs a Copilot and not everything needs a chatbot, but little strategic uses of AI within an application can really make it a user's experience a lot better. And I think developers need to look at the full spectrum of things that, that GPTs can offer.

As the vice president of developer communities, you must talk to a lot of developers just feeling overwhelmed. Where, where should they start?

I think that's a great point. People are overwhelmed, people are worried about their jobs, people are worried about where they fit. And I think that you, you are smarter than the AI. We need to remember that this is an artificial intelligence. We are creative, we are intelligent, and we are in the driver's seat. So it's not going to be AI that takes your job, it's going to be some nameless, faceless suit out there in the world. And what we need to do is figure out how we can use AI to make our jobs better, easier, more fun, how we can reduce toil. That word toil is one that I've really been focusing on. Which parts of my job are boring and tedious? Before AI, I would automate a lot of those things by writing batch files and PowerShell scripts. Now I'm using AI to write batch files and PowerShell scripts to automate the most boring parts of my job so I can focus on the fun, creative stuff.

The role is constantly changing for developers. What is— are some practical things they can do to infuse AI?

Well, you mentioned how do they educate themselves, and I didn't answer that well because I feel like people need to know that there's a ton of free courseware. Like, if you go to developer.microsoft.com/ai, right on the top there, there's 3 main links that point to MIT-licensed open-source GitHub repositories with full courses. You can do a course in a weekend, or you can have a full 26-week course that you could teach at the university level.. And I think understanding at some depth how these things work. If you treat AI as a scary black box that you send strings to and strings come back, I think you're gonna be unhappy as a developer. But if you understand what's really going on at the math level beneath, it'll help you better contextualize AI in your life and how you wanna fit it into your work.

So it really fits into knowing that one level deeper you always talk about?

Yeah.

How can you ensure developers don't skip that step with a lot of things they do because those LLMs are way too smart?

That is a great point. They do feel smart. It is easy to be amazed when a parrot talks back and you go, "Wow, that parrot speaks English. That's amazing." But the parrot is only parroting back what it hears around them. For us to really understand that AIs learn from your data and they are nothing without your data. If your company does not have your data organized in a useful fashion,, then your AI is gonna produce garbage. And that's why remembering that the data that you have, the information that you have as a person, or the information that your company has collected in data and data lakes makes the AI better. Understanding that relationship and learning about that is gonna make you a lot more effective as a developer.

Thank you so much for that developer angle for this keynote. Is there one last thing you— what do we really need to start doing as developers?

I think we should also spend some time thinking about ethics and responsibility, and we should spend time looking at the work that real thinkers around sociology and psychology are thinking about AI. How we interact with them, whether or not we pretend that they are people, or whether we realize that they are in fact tools. So don't just spend time teaching yourself the technical aspects of AI, also think about the ethical and the responsible AI. Things, and we've got resources on those as well.

That's a very good point. Thank you so much.

Thank you.

We just learned why AI is a must-learn for every developer and how it's reshaping the industry. But how did we get there? To understand the rise of AI, we need to talk about Python, the language at the heart of it all. And who better to guide us than the man who created Python himself, Guido van Rossum. Have you ever envisioned that Python would be as big as it is right now with all everything how it's used in the AI?

Never.

I'm— every time I realize it's grown another 10-20%, I'm surprised.

Were there anything particular for you that were built at that time that you thought it could become something big?

I didn't have any sort of major examples of things that had become big that I wanted to follow. I mean, remember, I started it in 1990, and the word open source had not been invented. There were, like, a few things that were released under a license that is now recognized as open source. And I I did take that as an example, but just as an example for the license, not as an example for how successful I wanted Python to be. I just wanted to share it and I wanted to have fun developing it.

It's nothing small anymore, the language you created at that time. We see many different libraries like PyTorch and Skykit and TensorFlow and so on. Did you ever think about how that really changed the AI world?

No, the AI angle was a complete surprise to me. When I started Python in the early '90s, it was still in the middle of the AI winter. So AI was not on my mind at all. Although my main focus was still on professionals. I think the first tagline I ever had was something like, "Fills the gap between C and shell." Was there anything particular that you tried to focus on development to get it more accessible to everyone? Making the language more accessible were a few philosophy bits, a few bits of philosophy that I inherited from another language. And that was about making programming less complex less complicated by, for example, using indentation instead of curly braces or begin-end blocks.

How did you get all that feedback? Was it from a couple of people, or was there more people involved to get it to where it is right now?

Oh, there's been a huge community. As soon as we open-sourced it, my inbox started overflowing with feedback from people who sort of picked it up, managed to build it, and immediately saw the potential for their use cases.

Was there one particular package that you think of that's like, okay, this really changed Python into a data science language?

I think that must have been NumPy, or its predecessor Numeric Python, because that sort of suddenly gave Python an ability that it didn't natively have, which was sort of efficient representation of large arrays of numbers. For me, that sort of one big moment was when TensorFlow came out, which was produced by a famous Googler, Jeff Dean. And when I worked at Google, he had never given me the time of day because he was so important at Google, and I was just a low individual contributor. And I didn't think that he or any of the other big Googlers cared much about Python because it was just a scripting language and all the important work was done in C++. So I was quite surprised and also felt vindicated that TensorFlow, even though it's still written in C++ and for good reasons, the user interface is actually you write scripts in Python.

So that was a proud moment.

That was a very proud moment when I saw that the first line of code on the TensorFlow docs was import TensorFlow.

I can imagine that.

Yeah.

Now you've created— it started with something small, became quite big to what it is right now. There's many people who would like to follow your steps. What advice would you be able to give someone who wants to create their own language or even want to create just another application?

If you want to create an application, go for it. For people who want to create their own language, the standard recommendation is go take some rest and wait for the feeling to pass, because that's sort of— it's unlikely to be successful, and even if it's not, it'll consume your life. On the other hand, if the feeling doesn't pass, Go ahead and learn everything you need to learn from looking at other languages, how they're implemented, how they're used, what their syntax flaws are, because language designers all steal from each other.

Python is still your little baby, I would say. It's not a little one, I guess.

Grown into Godzilla.

Well, and it will always follow you. We saw the number plate of your car. It also says Python, so it's, it's really your thing. We're very happy with what you've set up at that time, and we thank you for your time, right, today also.

Of course.

Thank you so much. Python's simplicity and power has made AI accessible to developers worldwide, and it's the foundation of many of the breakthroughs we see today. Now, one of the most exciting areas of AI is the rise of intelligent agents, AI systems that can think, act, and learn in real time. Now, to break this down, let's hear from Jennifer Marsman. She's the Principal AI Engineer in the Office of the CTO at Microsoft. Jennifer, could you please explain to us what AI agents are and what they are doing?

Yes. So an AI agent is simply a software process that you can delegate a chunk of work to, much like you could to a human colleague. So if you think about scenarios like customer service and software development. We're seeing a lot of that where people are able to utilize agents for help and augment their own capabilities with this work. But we're still very much in the early days of agents. So we believe that in order for agents to reach their true potential, we need a more robust infrastructure surrounding them. So not only the core foundation models, but 3 key things, and that is memory, entitlements, and actions. So let me break those things down. By memory, I mean that these agents should have a context of not only the information that they need to solve the problem, but also information about previous interactions with the user. And that can lead to a more personalized and relevant experience. Then there's entitlements. Now entitlements mean permissions. So if an agent is doing something on behalf of me, it should only have access to the things that I have read access to, and it should be able to only take actions that I have permissions to take. And then finally, there is the idea of actions. So if you have an agent that is, for instance, booking a restaurant reservation for you, you need to go beyond just the natural language capabilities of the model and actually call out to code. And so that's what, you know, tool use enables you to do with these agents.

Because of what new enhancements that we see or developments in the AI world are agents now so popular?

Well, I think that the capabilities of these foundation models continue to grow and expand and become even more powerful. And so What we see as a result of that is that now these foundation models are able to handle more complex, sophisticated tasks. And we're calling those reasoning models. And so the— what you see these agents now able to do is instead of just calling a model and getting a response back within a few seconds, the model is given a little bit more time to ruminate on an answer and really walk through a process of chain of thought reasoning and then be able to come up with answers to more complex questions. So it's empowered a whole new category of abilities to these agents.

You told us before already that you work with Fortune 100 companies in the Office of the CTO. Any use case that you're really surprised of, of where AI agents were used?

The very— the most exciting thing to me is actually the diversity of use cases that we've been able to do. There's been a number of things from looking at ideas like fraud detection and financial services, to looking at the entertainment industry and things like planning vacations and things like that, to financial services, to healthcare, to retail and shopping. So there's really been a wide variety of use cases that we've seen.

How do you see the future of the use of AI agents in the day-to-day life?

Well, I actually had one of the most magical moments with a, uh, with one of the reasoning models when I was doing early testing on a reasoning model, the O1 preview. Uh, the scenario that I tried actually something that I was doing for myself in my, in my personal life. I have 3 children and they're all in different grades. One is in middle school, high school, and elementary school. And I was trying to sign them up for camps for the summer. And in my area, the camps are very complex. And so what I did is I actually wrote a prompt that described my 3 kids.

Here are my 3 kids and what grade they're in and their interests.

And then I threw that whole catalog, like the text of all of those summer camps, at the prompt and said, help me figure out, like, write me a matrix of each week of the summer by each kid and what camp they should be in. And it did it. And that was just this amazing moment because that is a very, very complex constraint satisfaction problem where it had to keep track of a lot of nuanced detail, and the model was able to to do that, and that was really exciting to see.

So we already see a lot of developers trying to build AI agents. What advice could you give them to bring it to the next level?

Well, there's two things I would say. The first thing is, instead of focusing on what an agent is or is not, really focus on what is the value that an agent could bring to you. What are the things that you could automate away, and/or the product value that you can give to your customers or the way you can best serve others by creating a better service experience. I don't think we've actually hit the bottom of everything that we can do with just GPT-4o. These foundation models have such amazing capabilities. The more you dig, the more things you play with, the more experimenting you do, the more you find like new capabilities that you wouldn't think they could do.

I think the word experimenting is probably the most important one as the advice.

It really is.

Yeah, it really is.

Thank you very much, Jennifer, for that advice.

Yes, keep playing.

AI agents are transforming how we interact with technology, enabling smarter automation and more intuitive user experiences. But as AI systems grow more powerful, we must also ensure they remain fair, transparent, responsible. To explore this critical topic, let's hear from Sarah Bird. Hi Sarah, you lead the Responsible AI initiative at Microsoft. We start with a big question: why do you think responsible AI is such a critical element now in the, in the entire AI landscape?

You say it's a big question, but I feel like it's actually a really easy question because who wants to use an AI system that they don't trust, right? We see the potential of AI to change how we do everything, right? How we communicate with other people, how we work, how we have fun. But those are important areas of our life. We're not going to want to use AI for that if we can't trust it to behave in the way we want, aligned with the things we care about. We need to make sure that that it's secure, that it's safe, that it keeps our information private. And so I don't see how we get the exciting AI future if we don't figure out the responsible AI piece. We have to have the critical tools and technologies to make sure that we can build AI systems that we can control and that people trust with, you know, good reason.

Of course, developers building these AI systems face many challenges. Do you have any insight in how to implement those amazing AI features responsibly in their systems?

Yeah, I think first let me acknowledge that there are many challenges with generative AI, and not only many of them, but a really wide range. When this technology came onto the scene, we of course, you know, cared about things that we've been talking about in responsible AI for a while, like is the system Is it fair? Is it stereotyping? But we also saw new types of risk, like does it produce copyright material that you don't want to be using in your downstream application? It's— the systems are exposed to new types of attacks, like prompt injection attacks. And so with this much bigger range of things to be concerned about— and it's growing every day, in fact. We are seeing, you know, some of the newer technology we have to be concerned. Someday, is this going to be able to produce formulas for chemical weapons? Is this something that can cause much bigger risk? And so with this kind of growing set of risk, it's really important that we have a plan when we're building an AI system to be able to both understand what risks are in the system, be able to build a system that is safe by design, and then also to be able to test that systems so that you know that those mitigations you built in are actually effective. And then after you tested it and you feel like, okay, it's working, and you want to ship it, you also need to monitor and figure out what's going on, right? Adversaries find new ways to use systems. Users find exciting new ways to use systems that maybe you didn't plan for, and so it's not behaving how you expected. And so, uh, we really look at this complete playbook, and that's something that we expect every developer is going to need to do if they're building with AI.

And it's such a valuable insight because that has been the case, of course, with software even without AI features that we should really test and monitor and keep monitoring. And sometimes we forget with these cool new things that we also need to do these important aspects of AI.

I actually love that you said that because I regularly joke when I'm talking to people that, you know, the novel thing we've been doing to ship AI is test it, right? And obviously anyone in software engineering should be laughing because it's not novel that we need to test systems. But to be honest, uh, very few people who were building on top of AI were really testing at the level that's needed. And certainly when generative AI came onto the scene, we found that you need a lot more testing because It's so powerful and it's so flexible, so you have a lot of control over how it behaves in terms of how you configure the metaprompt, how you configure safety systems, how you feed the data into the system. And so with all of that power and flexibility, it can behave very, very differently. And so you have to test it and you have to fine-tune it to get to something that is really what you want to ship.

Developers, of course, still learning how to develop these systems, how to implement AI. How do you suppose they can juggle the constant having to or able to be injecting new AI features with the other side and doing that responsibly in respect to ethics and constantly being under pressure of the business?

Yeah, I think this question is often posed as a trade-off, but if we go back kind of to what I was saying at the beginning, I don't think there is a real trade-off here because if you're shipping a system without responsible AI, most people aren't even going to be comfortable to use it. And so, you know, at Microsoft, we very much in our own practice of this view it as just part of shipping the AI system, right? That it's not complete until we've done the appropriate testing, until it's working the way that,, you know, we actually want it to behave. And so we've kind of gotten out of the mindset that this is a trade-off. Now we recognize that the market's moving really quickly, and we also want to innovate very quickly. And so a lot of our investment is figuring out how we do all of the responsible AI practices at the speed in which AI is moving. And for example, one of the things we've built is what you call kind of automated red teaming, or, you know, we call it safety evaluations. But this is taking the tests that we think are important and using AI to run those tests against systems and scale that up. And so we've got a robust testing system that just every team in the company can run. And so they don't all have to spend time investing in their own testing. They can start from what we've already built and customize it. And so a lot of our approach of making is not a trade-off, is really figuring out how to build the right tools to scale it up, to make sure that everyone can use the same tool, and make sure we're just getting it right in one place and that people can build on top of that. And so we think, you know, the same thing for our customers or for developers. I'm lucky to be at Microsoft where we might have, you know, more responsible AI experts than many, probably any organization in the in the world, and that's a great privilege. And so we can invest a lot in this space, but not every organization can do that. And so our job is to build tools and technologies that other people can just use and build on top of so that you don't have to start from scratch, so that it doesn't need to be a trade-off.

What would be your one tip for the developer?

Test your systems, right? We've always known that the best software practice is to test systems. And with AI, you have to be more thoughtful and test for a wider range of things. But it's still the same thing. Test your systems.

So true. Thank you so much for being here and sharing all these valuable insights with us.

And it was great having you. Yeah, thank you so much for having me.

I'm excited to see what everyone builds.

We've just heard from some of the top minds in AI about why this technology is transforming the way we build software.— how Python has been foundational to its growth and the latest advancements shaping the future. But here's the thing: learning about AI isn't just for experts. It's for every developer, including you.

That's why we're so excited for you to take the next step at your local bootcamp. This is your chance to get hands-on, experiment with AI, collaborate with peers, and start building. Whether you're brand new to AI or already coding with it, there's something for everyone.

We also want to take a moment to say a huge thank you to all the local organizers who have made these bootcamps possible. Your dedication and effort are what makes this community so incredible.

And don't forget, make sure to request your badge and register on globalai.community. This is your gateway to staying connected, accessing more learning resources, and being part of a growing global network of AI enthusiasts and developers.

The future of AI isn't just being built in research labs. It's being built by developers like you in communities all around the world. This bootcamp is your opportunity to explore, create, and be part of something bigger.

So let's get started. Dive in, ask questions, build something amazing, and most importantly, have fun. We can't wait to see what you create.

The next AI breakthrough could start right here with you. So join the community, claim your badge, and let's go!

Let's go. I'm so excited. That's amazing. I hope you liked it. That's pretty cool. Hello, what's up, Sherry? I couldn't see you because the thing, but it's good to see your face. You all, of course, are welcome to join us on camera if you'd like to. We are here live at Miami Dade College. So anyone in the area, you can all come on down if you want, but we'll be doing both online, offline. Bunch of people joined since the session began. So hello, Nan and Sherry. Good to see you.

All.

Let's see if I can find more people that I haven't already called out. Sam, good to see you. Um, let's see here. Tony, what's up? Uh, let's see, I'm going through the list. I feel like I've called— maybe I'll recall you out. Derek, good to see you, my friend. Afua, of course I mentioned you all as well. Um, but we have about 45 people who have joined us online and a, a whole room full of people. Alonzo is in the house. So thanks for being here. Amazing. So just, I guess we need to start first by just understanding where you all are at from an LLM perspective. Most of you have been, like, part of my community and building with me for quite some time. But some of you are brand new to me. So I would love for you on a scale of 1 to 10 to kinda state your, your feeling about, like, technical aptitude in the world of AI. Like, could you build a system? Could you create something? Have you ever used an LLM? On 1 to 10, put in the chat your number of how good you feel about this. I know some— it's okay. You could be humble or not humble. It's up to you. But all sorts of numbers should be in here because I know who many of you are. But just throw it in the chat. 1 to 10. What do you think? Someone's got to start it. All right, awesome, awesome. All right, let me see it, let me see it. Um, hold on, there we go, 37. All right, I like it, I like it. 842. All right, cool. So we got some ones, some new people, which is awesome because this is the time to learn this skill. Also some mid-range. Most of us are pretty humble in our approach, not Eric, but, uh, really awesome. Great. Okay, awesome. Rob, I see you. Tony, amazing. All right. So this is great. We've got a really good wide range. So here's the good news. Take a look at these numbers. So we're all friends. Many of us are very eager and interested in helping everyone kind of get elevated in this skill set. So if you see someone with a high number, like, take note, become friends, reach out to them on LinkedIn. It's a great opportunity to just increase your network. Because— and, and of course, also know you're not alone. There's not a single number in here that's all by itself. So there you can be anywhere on this journey, and where you will be a week from now is different than where you are today. And this is a big part of that journey. So I'm super excited to have you all here. Thank you for being here. I will approve your bots if you want to, to record this. You're welcome to do it. If I don't approve it right away, it's just because I haven't seen it yet. But great to see you all. If you know, now is your chance. We're gonna spend a couple of hours together here. And if you want to right now, now's your chance to go ahead and share the link out to people. It's great if you share the registration link. So I'll put that in the chat. But you can also share it right to Zoom if you want to. The problem with that is they won't get the replay, they won't get the keynote, they won't get the badge. So I encourage you to send them that link. And Eric started the fun fanfare of connecting in real life with real people. So please go ahead and throw in your LinkedIn if you would encourage people to connect with you. I do it for sure. I think I'm connected with all of you. That's why you're here. But if we're not friends, let's be friends. And definitely, yes, awesome, Mauricio. Thank you for that. Oh, good. Yes. Allow, allow. All right. So as the bots come in, we don't have that many. So I was looking at our numbers. I'm like, oh, man, maybe half of these are robots, but they're not. But I don't mind. I love robots. I love you all. You're all welcome here. Robots and humans alike. So welcome. What's up, Adam? Are you in a car? Is that what's happening? I can see the light. What happens to me when I'm in a vehicle? Awesome. But he is not driving. Don't worry, everyone. Amazing. You— oh, okay. I thought I am recording, but I'll record again. So this will be recording locally. Thank you for that suggestion. But I'm also— this is streaming on all places. So we are on LinkedIn, Facebook, Twitch, you name the platform. We are streaming there, and that means it will be recorded in the cloud in multiple locations. You won't miss out. So the first thing I wanna do is now that I know kind of where some of you are, I'm gonna ask for kind of empathy. Many of us remember what it's like to be a one in any number of topics and how intimidating this world can be. And many of us have a lot of experience. So I'm gonna encourage you as we go through, we're gonna have lots of time where we're gonna talk about a thing and then we're gonna do a thing. And during that doing, feel free to share. We also have an opportunity for some lightning talks at the end. So if any of you are interested in actually standing up— well, in your case, getting up in front of everyone, sharing a little bit about what you do, your journey, your interest in this field, what keeps you motivated, any— your product. I mean, do we have a product launch that's happening, Eric? Oh no, I'm just very interested. We're waiting very— we're all here. We're all here for it. Like, I went in and I didn't even matter. I looked at like some fake pricing, and I'm like, give it to me, I'll pay you right back. So really excited about what some of us in the community are doing and would love if you have it in you and you are interested in sharing a little bit about yourself. We're talking 7 to 10 minutes, nothing crazy. But you are welcome to do that. So just send me a message direct and I will line you up. We have about 4 people right now, but I would love to have more. It's all about— today is all about community, global AI community. Like, we are all learning this skill set. We're all at different levels. We're all coming from it from very different angles. So now comes that time, where I also wanna find out where you are coming from from an industry perspective. So if you'd be up for it, why don't you throw into the chat the industry you're coming from? And you could be cross-industry. That's fine. You could put in cross-industry if you're a consultant or something like that. But if you're coming from healthcare— oh, good. Financial services is in the house. You know, I'll let you all throw it in there, but that'll be super helpful so we can kind of It's good for all of us to know where we're all at because those of us in healthcare can bound together. All right, there we go. What agriculture? Oh, oh, and AG lending. What's AG? Is that like— I thought it was agriculture, but I don't think that's right. Healthcare, manufacturing, amazing. Yes, AI app development, platform development, which is actually a distinction because you're not just building an app for people. Consultants, love it. Power generation, youth soccer. Do you know that Warren runs a soccer program? Do you know that he runs Precision Soccer? It's like data and soccer. Anyway, you should reach out to him, or I can connect you guys. I know you know him at least through our events, but I don't know if you all— if you knew that. So let's see, nonprofit, yes. Age tech, HR analytics, cross-industry, awesome. So we have so many different wide variety of different industries. And this technology, once you understand the technology, becomes this huge lever that you can pull in your specific domain. I actually think what, you know, a couple, many of the speakers, I mean, I met, I've worked with every single speaker that was in that keynote. And Jennifer Marsman, she's an architect. We both worked on applied AI kind of in the very beginning. She was running a project called AI for Earth. I don't know if y'all remember it. You could Google some of those YouTube videos. Fascinating. But she was the evangelist for the AI for Earth platform, which was all about, like, sustainability and using AI in agriculture and, like, food sourcing and power and energy and all of this. So really cool work that she got to do. Now she's working kind of more broadly. But, but during that time, we were both very passionate about, like, demonstrations, like building actual demos. And You know, you work in these companies, most people are kind of afraid of demos. They're like, ah, it's gonna break and it won't work. And, and she and I shared a common, like, live on the edge of danger. So that's why I really enjoyed working with her. But we have a wide variety of, of different industries. And what I learned really working with Jennifer and Sarah Bird, who runs the Responsible AI product, I really realized that it is in the doing, it is in the practical application that this technology becomes the most powerful. Like, it's, it's powerful to a degree in like general, generalist usage, but it becomes extremely powerful when you apply it to agriculture or, and not just financial services, right? But banking or insurance or like very specific application. And that's where your lens, like whatever you've worked on, your lived experience really does become a powerhouse for you to use. So what I'm gonna do now is jot down my dream agenda for today, what we're gonna cover in the next few hours. And then I will also share with you the lab exercises that we're going to do. And then, and then, yeah, and then we'll, we'll see how it goes. Let me see how this marker works. It looks new. It's purple. I think I might have left this here last time I was here. But I'm gonna start with— let's see if you all— oh, no, no, no, that won't work. Let's see, we got green. Let's see if green works out a little bit better. Oh, dear. Alright, we'll see what happens. Alright. So now, instead, I'm gonna— let me see if I got another color. I shouldn't write in Sharpie. That would be bad, right? That'd be bad. Let me see. Oh, I got a black one.

Let's see.

It's not terrible. We'll go with that one. We'll go with that one. Alright.

Okay.

So agenda part number 1 is we are going to have to talk about an introduction to— you could kinda see it— LLMs. Right? So I'm gonna take you through, and again, it's not my content. This is part of what the Global AI Bootcamp would like me to convey with you. I have my own perspectives on many of these topics, but I'm going to try Y'all know me. I'm gonna try and skip to, to stick to the script. It's hard. So I'm gonna stick to as much of the script as I can. But I do want to share with you some more practical perspectives than what we might be talking about. So it's a little bit academic, but then I'll throw in Noelle's perspectives. So the first thing we're gonna do is talk about LLMs. That's kind of the core AI bootcamp content for today. We then will go in and take a look at how to leverage some baseline models in a product known as AI Builder. And some of you were in our AI Builder Live Learning Lab last month. So you will have seen this, but many of you haven't. And I've got Azure credits for a few of you. So if you need Azure credits, I don't have, like, enough for everybody. So if you have an Azure account, please use it. If you have my account, use AI Leadership Institute. But if you're brand new and you've never used Azure before, I can set you up with some free credits for you to go to use. After we talk about large language models, I then wanna talk about something that Sarah mentioned, which is the concept of AI. Oh gosh, I always spell this wrong. Anyway, there's an E in there somewhere, but AI safety, right? So how do we build AI safety systems? And then the Last one is, I'm gonna call it what the world is calling it right now, vibe coding. I mean, like, what is a world that, like, creates a word that sounds so much like Noelle? But vibe coding is what we're gonna talk about, how to get started. So this is like an intro to LLMs, an intro to AI safety, and an intro to vibe coding. That is today's agenda. We will have labs for each one of these. However, these labs, because of the nature of how we are working, I'm gonna tell you what to do, and then you're gonna go do it. And I'm just gonna wish for you the best. That's why knowing who's in this list is super important, because they can help you. We'll also be taking questions if you get stuck. Everyone here can unmute themselves. I'd encourage you not to do it while we're trying to, like, share a message. But during the Q&A period, during the lab period, you can unmute yourself and be like, I have a question. This isn't working. I will just share with you candidly that often— oops, I forgot to do this for my poor recording people. Oh, yeah, look. Okay, let me try this. Okay, there. Just in case the recording people— we have a couple things that are recording right now, so I'm trying to create better audio for them. It's a little late now. Oopsie. Okay, but what I think would be one of the most common challenges we have is not about the tech. It's actually administrative, right? Salim is here. So administrative, logging in, authenticating against a tenant like this just eats up all of the joy in our lives. But I'm just sharing that with you. So if you run into that problem, it's okay. It might take a minute to fix it. So just try and hopefully stay a little present in what we're teaching, and then we can work with you to get it done. Sometimes I can't fix it in a day because you have to like— and I might ask you to do crazy Microsoft things like reboot your computer and clear your cookies and shake your keys over your keyboard because that often will work to solve these problems. But many of us, I would say 50% of the people on this call have gone through this process with people before. And so you are not alone. But just realize like that first step of just getting access to the environment can be kind of the hardest thing we do today, which is unfortunate, but it's security. We'll deal with it. But I just wanted to let you know that that is, that is the case. So let's get started first with LLMs. But I also need to plug in my computer. So hold on one second. Thank you. Okay. Looks like— all right, perfect.

We're in.

There we go. Ready for Lifelong Support. All right, so I'm going to share my screen and share with you a presentation that they created for us, but like I said, I have my own take, so I'll be presenting you the slides. You will have access to these slides. So if you like what's on there, I'm also going to encourage you to kind of see this through the lens of something that you could potentially offer one day. I know right now you're like, I just learned this stuff. But the power of amplification of the world of AI really comes in as soon as you learn something, teaching it to someone else. There's two benefits. One, you learn it more when you teach it to somebody. And two, other people need to know this that will only maybe come to you to learn it. So I think many of us underestimate the impact we can have sharing our ideas with the world, and these skills are really, really important. All right, awesome.

Okay, here we go.

So let me share my screen again. I don't think there's— oh, I wonder, I might do a couple things. All right, I'm gonna share my screen and see if I can find my PowerPoint.

Presentation.

It says I can share sound. I don't know if that's true, but I'm gonna try it. Okay, I don't really know which one of these is the right one. I'm gonna go with this one. Let's just see what happens. All right. Uh-oh. Oh yeah, this is it. This is it. All right, cool. All right, so this is one of the first ones, and it's basically— oh wait, maybe not.

Oops.

Hold on, let me stop. Wait, let me see if I can just open up a new one and you all see it. Open recent. Gosh, AI agents, LLMs. It should be called, maybe it's generative AI. Can you see it says getting started with generative AI? Can you see that? Right now there's like, oh gosh. Let me see. Oh no, this is not what I wanted. Oh well, I'm going to start with the other one. Let's see. Okay, cool. Thank you. Thank you. Let me see here.

Uh-uh-uh.

I gotta scroll. Oh no. Okay, cool. So let me go back. I'm gonna stop sharing and reshare the right presentation. All right. I think I can't really see what these are. That's Agent Service. AI agents.

I think it's Is this one?

All right, I'm gonna go to this one. So this, of course, was created by Microsoft MVPs. So I should share with you a little bit about that history. What's up? Adam's in the house for reals. Thanks for coming, man. You didn't have to. I was just asking. But he's here. Yay. See, join us. All right, awesome. So the Microsoft MVP program is a process of application. So you basically apply or someone really nominates you as an industry professional. You You get nominated to Microsoft as a, my, most valuable professional. In that nomination, what validates you as an MVP is how much work you're doing in the community to share the ideas that Microsoft cares about, whether it's core values of Microsoft, whether it's AI, whether it's technology platform. It could be anything across all of Microsoft's ecosystem. I am an MVP that has been awarded this. Will hopefully be my 5th year. So big year. My 5th year in AI services, meaning the engineering side of Microsoft AI, as well as Trustworthy AI, which is the responsible AI platform. Now, you could actually— I mean, I, I was just at MVP Summit last week. So that's one of the benefits is every year we come together. And all the MVPs, sometimes it's the only time I see my friend, my other friend MVPs. But we come together once a year, and the Microsoft teams give us inside scoops on what's coming up. In the next year, mainly so that they can create in us evangelists for new products and services that are coming out in the coming year. We are under NDA, so we're not allowed to share any of the cool new stuff until the day it launches. But we get a book of all those launch dates, so we get to train ourselves, get up to speed, build, you know, solutions. And that way on day of launch, we're already kind of pre-configured to understand the technology and be able to help people who are trying to implement it. As soon as it goes live. And many times because of the concepts of like private preview and public preview and then GA, we are often invited to start using it in private preview, which means only the top clients of Microsoft as well as the MVPs get access. So that's how I got early access to working during COVID I got a chance to work on Codex with OpenAI as part of a new, at that time, new relationship that Microsoft was building with OpenAI. Codex eventually became GitHub Copilot and ended up being a model that was used inside of ChatGPT.

But none—

I didn't know any of that. I was just an, you know, an NDA-governed, you know, model tester. So as you heard Sarah Bird say in the keynote presentation, testing is everything. And so I always said yes to opportunities to test a model. That's why, like, Eric recently, was it in January or maybe December? I can't remember. But came to us and was like, hey, I've got a new platform. And so he kind of taught us all his platform, and I kind of looked at it through the lens of testing. Like, I love testing. If somebody wants me to test out their model, I'm like, let's go. I love that. It's a great way to learn the technology, but it's also a really great way to provide feedback to ensure that that model is created, I don't know, in the most fair and just way. So that being said, I will. Oh, let me turn this into presentation mode. Hopefully it will stay shared. Okay, it's thinking. There we go. All right, Alonzo, what do you see? Do you see this? Oh, it doesn't say Azure. Show that on the right side. Oh, yeah, yeah, yeah. Okay, cool. So that means I just need to swap displays. I can find my mouse. Oh, where's my mouse? Oh my goodness! Oh, well, at least the AI is translating my speech, so that's exciting. Anyone see my mouse? Oh dear! All right, escape.

Yeah, it's—

oh my goodness! All right, hold on. There he is. I found it. Okay, let me try that again. This time, let me try it one more time. Try and keep an eye on my mouse the whole time. There we go. Swap displays. Better. Alright, here we go. Are you ready? Alright. So I didn't go and change all the speaker names, like, because it would be my name everywhere. So you seem kind of silly, but I did add some cool stuff to some future, some of the presentations we'll use later. But again, the reason I show this to you is because this is what you get when you become an organizer of a boot camp. They give you this presentation.. And then you just have to use the information in your brain and in the lived experience you have in the world of AI to add a little bit of color. The reason I like this presentation, and the reason I chose it, there's a bunch that I could have chosen, was that I thought it would be good to level set you on all the things that Microsoft is doing in the world of AI. And of course, I worked for Microsoft in the time that was building these models at the very beginning. I've come to call or refer to this time in my experience at Microsoft, baby tiger mode, where we were building all these tiny little models with no real understanding of how big they would get or how, what they would do or who they would serve. And so now as we've, you know, kind of evolved, many of those models are now doing big things for billions of clients. And so, or billions of customer requests.. And so I think it's even more important now to understand kind of where things started. So of course, this is, I think, probably one of the most, interesting, you know, diagrams. You've probably seen it over time, right? Generative AI is the fastest growing technology. It's achieved the hype cycle of no other product, even the most crazy products that came before from mobile phones, which took a long time, Some people are still getting adapted to mobile phones. Um, the internet, uh, Facebook, Twitter, um, TikTok, like all those, we kept adopting technology faster and faster. But to get to 100 million users, 3 months, like that's crazy. Uh, it took Alexa, for example, 5 years to get to 100 million users and now they're over 500 million. But still, I mean, we're way past that now. But it just gives you a sense of kind of how quickly this has been adapted. Now, this is a timeline. I like to talk about this because I was there for a couple of different pieces. Obviously not in 1991, but I do like to represent one of the most important steps in Microsoft's AI journey is that it started with research, as all AI does. It starts in research. So Microsoft Research was established. They started doing cool stuff. And the— where I stepped into the journey was in 2014. I remember, for example, joining Amazon Alexa. And I don't know if you remember, but back then there was like some swag. And I had a sweatshirt and it had a blue ring on it. And I did not know at that time that Cortana's logo was also a blue ring. But Amazon Alexa kind of took it and made it slightly different but almost the same. So people would say in the elevators that I was going to events, they'd be like, hey, is that Cortana? And I'd be like, huh, no, it's Alexa. Like, I don't even know what Cortana is. Um, but Cortana and voice recognition came out the exact same year as Alexa, which most people don't realize. Cortana died a slow death. Alexa, well, it's still kind of— the cards are still being played, but It's, it's still live and in many people's houses. The Responsible AI strategy was developed. The team was created in 2020. But none of this is, like, new. These are when, like, moments were established where organizations were created, but the efforts were longstanding. And then, of course, last year, or I guess now 2 years ago, Copilot was announced. But this is not their entire journey. It's just some key points to kinda let you know. I think the most important thing on this slide is that It all started with research. And therefore, I mean, I can extrapolate that to every project I've ever built starts in kind of a research phase, a hypothesis test. And that is a good perspective to have if you're new to the world of AI, that everything you do in AI starts with a similar hypothesis test. So there are a couple cool things that are happening, in the world of artificial intelligence. I'm going to share a few things with you, but before I do that, I wanna share with you kind of the architecture of Microsoft Azure. There's a lot to it. I'm gonna then go into AI Builder and show you some of these tools. But when I started with Microsoft again in like, I was at Amazon Alexa from 2014 until 2018. And then I went at, in 2018 to Microsoft to help launch a lot of these services that you see on this screen. And what I, I was hired into research to help organizations, like the research organization productize artificial intelligence. In other words, take it from becoming an academic project, like something you'd see at Stanford or MIT, and turning it into an Azure service running inside of Azure. So I was going from Harry Shum's organization at Microsoft, which is head of research, to Scott Guthrie's organization, who's head of all of Azure. And it's a very tough road because you're going from, hey, let's just test stuff out on, you know, and have fun and do research and then get a PhD to we now have to provide secure, scalable services that are indemnified legally to support businesses. So now all of a sudden, when something goes down, it actually costs the organization money. And that is a whole new world. Most academics don't think in these terms. So just to give you a sense of some of the models, I like to think of it as an umbrella. There's two different sides. There's what's called applied AI, which is the applied AI piece. Those are all the models that already exist. So these are models that have been created and pre-trained by Microsoft. That includes things like AI computer vision, AI speech recognition, translation, Oh, sorry. Yeah, I just mute.

No.

Yeah, you got it. And then machine translation. So in 2018, 2019, Azure's machine translation capability surpassed human translation. And what that means is that it reached what we called human parity. And that is the goal of most systems, right? It goes back to the original idea of the Turing test, right? In the '50s, Alan Turing created a concept that he shared with scientists to say, can you get a machine machine to pass a test with humans. And as a human interacts with that machine, does that human know it's a human or a machine? Can it tell the difference? Some of us know, as a matter of fact, I was doing this in the car ride on the way over where I would turn on, I was listening to some neural voices, and as soon as I heard the voice, I could tell it was a machine, right? In other words, they failed the Turing test. Like, as soon as I heard it, I was like, ah, This is not good. And but then there were other voices that sounded really human. And so in that case, it would pass the Turing test. So the fact that in 2018 these models were not only passing the Turing test, meaning that a human could not tell if it was being translated by a human or by a machine, but it accelerated that by actually surpassing the error rate. So that's the next thing. Once Turing test passed, that's good. You want someone to be you know, able to think that they're having a human interaction when it's a machine, you're winning. But the next step is how much error does having a human in that process cause? And can you create an AI system that eliminates that error to a higher degree? So if having a human involved, your error rate, like in autonomous driving, your human error rate is 75%. If you can create an autonomous vehicle or an AI system that has an 85% it's still gonna fail 15% of the time, but it's not gonna fail as much as if you had humans. And that becomes a huge accelerant for adoption of these types of models. So that happened in 2018, and there's a bunch of— a new timeline for that. But Azure AI Search, that was Semantic Search's release. So the ability to implement intelligent search across enterprises. It's funny. That was really, really hard to sell when I was selling AI services because people were like, ah, I don't know if I need to search for things faster. I'm not sure the value of that. But then enter generative AI, right? We started building LLMs, and those LLMs were searching for answers. All of a sudden, that semantic layer became huge. And so now AI search as part of an LLM implementation, many of you have heard the term retrieval augmented generation or RAG. That retrieval augmented generation architecture is based typically on a semantic search, Azure AI Search. Type of backend. Document intelligence. Many of us who have been in the AI Leadership Institute, you've done this. But being able to hand it a passport or a driver's license or any of, any kind of document and have it digitized, but also have the data analyzed and interpreted and insights driven. All of that comes out of— and now it's not just document intelligence, which means, can I tell what text OCR style, right? Optical character recognition. Can I tell what text is on the document? Now it's about content intelligence. Not only can I tell that there's text there, but what does the text actually mean? So I don't— if you've ever done this, like, right now on your smartphone, if you have an Apple, Apple has a new icon on every image that you take a picture of that has like— it looks like a little document, like a little piece of paper. When you click on it, it allows you to scrape the text from the image that you've taken a picture of. It's pretty cool. However, when you scrape that text, it doesn't know what it's scraping. So sometimes it'll put spaces in the wrong place, or it'll think something's a word even though it's numbers. Like, it's not very consistent because there's no content intelligence. So that's the future, is like, now we're helping companies, like, get their, their documents digitized. But in the future, meaning, while the tech is there now, we want to accelerate that into content intelligence and content awareness. And then that leads us to the second section we're gonna talk about today, which is AI safety. There's an entire suite of services that allow you to augment your AI red teaming capability, which everyone should have. But how are you testing your AI systems so that they don't hurt people once it's launched, so that they don't violate trust, they don't, share data they're not supposed to, that they don't exhibit unfairness or toxicity All these things are required to be tested. And now we've got AI systems that can amplify your ability to test, but it still requires a human to turn that on and monitor the output. Alright? So this is kind of the world of Azure AI. Now, I have delivered this session to executives. I mean, this is not new. These best-in-class services are now— we're going on 10 years old. However, I talked to you about the umbrella. One side is applied AI, prebuilt models. The other side is custom AI. And you could almost use the analogy of closed versus open. So applied AI are closed models for the most part. They end up being models that have already been pre-trained. You give it data, it gives you an answer. You don't wanna know the details. Code last, right? So it's more like a no-code, low-code environment. Then on this side, you have not applied, but now you have custom AI where you go in and you look at a model catalog of 2 million models. You pick one of those models and you custom train that model yourself. So those are kind of the two sides where I find a lot of joy is right in the middle, customizable AI, which means I take a prebuilt model and then I add to it additional data to get an output I want. Retrieval augmented generation is a common, like, customizable AI system where you can create— you can use an LLM, a prebuilt model, and then you can apply very specific prompting techniques and system prompting to customize how it answers the question. And add to that, of course, a dataset that you've configured and you've customized. So those are the 3 kind of architectures when you're looking at it. It starts with prebuilt AI and then completely build your own AI by grabbing a model and building it, like training it yourself. But then there's this magic in the middle where you take an existing model and customize it. You build on top of the training it already has. And that, I think, saves you a lot of time, but it's also much less expensive. Just to give you a sense on pricing, right? Like, if you're looking at dollar signs, The prebuilt models are super cheap to use, but don't have a lot of flexibility. Building a model from scratch is about, I don't know, $10 million, $30 million, $45 million, depends on how much data you need to have in order to get your right outcome and how much time you want to spend testing it. In the middle, you end up somewhere in the $2,000 to $15,000 range. Very achievable for organizations. And often where you choose to ask questions. So, and if you have questions, of course, don't forget you could throw them in the chat. I don't think I have it open right now, so I'm gonna close this out and open up our chat so I can see it. But if you have questions or comments, please feel free to chat with each other. I don't know where my chat is, so you'll have to talk it up. If you see a question and you have an opinion or an answer, please feel free to jump right in like you're not We're a community here. We're learning from each other. So feel free to do that. And maybe if one of you all see a question, Alonzo, in the chat, you can let me know. All right. So late recently, this is also an evolution, which I think is interesting to know. So first, when each one of these models, speech, voice, decision, translation, when each one of these models were created, they created their own AI platform, their own dedicated platform. So, like, speech, vision, custom vision, everyone has their own website you would go to, to do, to use the model. Then last year, we created an AI Studio, and the AI Studio brought all those independent websites into a single place. Then just recently last year, we launched— Microsoft launched AI Foundry. And Foundry is such a good term because it's basically where you go to build stuff. And so the Azure AI Foundry is the next generation of the aggregation of all of the different models you can use, both across prebuilt models as well as customizable models, as well as building models from scratch. So it's where you go depending on what you wanna do. So the most important thing you need to know when you go into Azure AI Foundry is where are you What are you trying to do? Right? Because what you're trying to do will dictate, do you go with a prebuilt model or do you need to start from scratch? I'd say you almost never need to start from scratch. But there are certain use cases, Bloomberg GPT and Mayo Clinic's GPT, like, they thought they were building something that had never been built before. My suggestion to you is to look through model, the model catalog. Look through the model catalog, do a search on GitHub, make sure that there isn't something you can start with because starting from scratch, is unnecessarily expensive. It also requires you to have a much different infrastructure investment than if you build on a model that's already been pre-trained.

Alright.

So this is what Satya Nadella spoke about at both Build last year as well as Ignite, which is their business user conference. Copilot, this is how Microsoft sees the world of applied artificial intelligence, artificial intelligence that is gonna be provided to a consumer where no technical skills are necessary. So we all have heard of Microsoft Copilot. You probably have used it. It's cute little rainbow type icon that shows up everywhere. We also now have 365 Copilot, which is a custom version that sits inside your tenant to help you with things. It still doesn't help me. I mean, here's the whole thing. Like, it helps a little bit. I like it. It's better than not having it. However, as soon as you start using a commoditized piece of software, like 365 Copilot, as soon as you start doing that, you're going to expose a gap, a gap between what it does and what you wish it would do. The difference is, is now with no relative— I mean, very little coding skills, you could now build your own gap-filling product. You can create your own Copilot. So this is called Copilot extensibility. You can now create agents that fill the gap between what Copilot does out of the box and what you wish it would do. For example, I really wish it would triage my email. Like, literally go into my email, read it, and tell me if I need to pay attention to it. I'm surprised we don't have this capability already. I mean, I have invested in all of them—Motion, Superhuman—none of them read emails. Remember what I mentioned about document intelligence? All of them know that an email came in. All of them know that it came from maybe someone important, but it doesn't have content intelligence. It doesn't know what's in the email. It doesn't know if it's a hi, hello, or a this invoice is late, or I'm gonna pay you a bunch of money. Like, I would like to know those distinctions so I can prioritize who I talk to, who I look at. If I get 600 emails, I might look at 100 of them. I wanna only look at the most important 100. And so that content intelligence is now— I've created that through. And of course, Microsoft will eventually get there. Like, the tech will catch up. It's what do you do in the gap, right? Mind the gap. What do you do when you've got the product? It's not where you want it to be yet. Microsoft decided to expose the ability to customize the stack. And that is huge, especially in the world of vibe coding where you don't actually have to know how to code to build the gap-filling technology. It's a really— you do have to be engineering-minded, but you don't have to know the language. You don't have to know how to build software. You don't have to know how to deploy software. You can actually leverage an AI system to guide you through those processes and those steps, as scary as it might sound. So really exciting. I'm a big believer in— I've been using a product known as Semantic Kernel for like 3 years now. I was one of the earliest users. I, I was huge. I tried to push it on like enterprise stacks and they weren't ready. They're ready a little bit now because we've coined the term agentic AI and we're teaching people how to build these like tiny little systems that talk to each other and orchestrate. 3 years ago, I was, this technology launched and it was so powerful, but most people couldn't do it. So I encourage you to look at Semantic Kernel. I also encourage you to go— Semantic Kernel has its own Discord. So, like, join the Discord. You can hop on with the developers. I mean, that's kind of the cool thing about being in this technology community is that we get to actually talk to the people writing the software. I mean, I think that's one of the coolest things that I get to do is talk to people that are making decisions about what to keep and what to throw away, where the the line is drawn on what makes a sprint, like being able to communicate with people and understand the decisions that they're making when it comes to product features, I think is pretty cool. So Semantic Kernel is one of those products, very much similar to how things are, are built in open source, but it's Microsoft, which is not normally known for its open source capabilities. So this is how it's implemented. You start with Azure AI Foundry. That is where the Copilot stack begins. And then Copilot Studio is where you build those gap-filling agents. There's about 20 maybe preexisting agents. I'll demonstrate that for you in just a second. There are 20 preexisting agents, and if you can use it, great. But you can also go very quickly and build your own agent. But when you start building agents, which we'll talk about today, when you start building agents, I think one of the most important things you have to remember is that an agent is just a single AI system. So you need to build that single AI system as if it was the most important thing you're doing. And then when you're done with that, you're never done, by the way. But when you're completed the process of developing it and you start testing it and you move on to the next system, you need to treat that next system again as a completely holistic, testable unit of work. And what most people are doing is massively creating agents that are kinda half done. And I'll just say it, Like, that's how Skynet is born, right? Like, that's literally how the end of the world comes, is us kind of haphazardly building a bunch of systems that then are gonna talk to each other and make autonomous decisions that we don't even understand. So our intentionality in building these systems is probably the most critical thing that you, you will focus on is how do I make sure that I've implemented this thing, this machine to do a specific task in a specific way that I monitor and have insight to at all times, and then duplicate that exact effort. Is it hard? Yes. Does it require lots of testing? Yes. Is it fun? Yes. So hopefully not terrible. All right. So I wanna take you through a little bit about LLMs, and then I'm gonna walk you through a couple demos. You're welcome to do those demos with me if you have an Azure AI, system. Otherwise, we're gonna take a break, and I will set you up. Those of you who need access, I will set you up with your Azure environment. So most people, again, there's a couple ones in here. So if you know how LLMs work, please just bear with me for a couple seconds while we talk to those who are new to the world of AI. Though you might know this as well. I always like to talk to people about the history, kind of how we got here. There was a, a, there was a challenge. And I was working when this was a challenge at Amazon Alexa. And the challenge with Amazon Alexa was that we were struggling. We were really struggling with the fact that we had built an intent-based system, which meant that when you said something to an Alexa device, Alexa would hear what you said, and then it would turn it into an intent, meaning it would interpret what you said and translate what you said into a thing that the machine could do. And it was a direct correlation, one-to-one. And so the challenge with that, of course, is if you said something that I could not correlate to an intent, it wouldn't work. Give you an example. I was working with a financial company. We have a couple financial, people here working with a financial company, and they wanted to create a voice-enabled transactional service so you could actually use your voice to transact money from one account to another, whether it was transfer or put into savings or any of these types of commands. The challenge we ran into was not everybody said, transfer $5 from savings to checking. Right? People would say all sorts of things like buckaroos or, uh, dinero. And not like because they were speaking a language, they would just use this contextually. We found out there were 7,000 ways that people would say over the course of 2 years, 7,000 different ways people would say transfer money from this account to this account. 7,000. That's ridiculous. Like humans, humans are crazy. We say all sorts of different things. Depends on how we're feeling. Sometimes we use more words, sometimes we use less words. So the challenge with that time was that we had to hard code 7,000 intents to catch, or 7,000 utterances is what we called them, to catch those intents. Every time someone said it a new way, we had to train the model to listen to that way and map it to the intent of transferring money. And that was a lot of work. It was mindless, mind-numbing work. And so We, the, the industry, Google mainly, was looking for a way. I mean, remember after Alexa was launched, 3 years later Google Assistant launched. Google ran into the exact same problem of intent mapping, and they're like, this is nonsense. What are we doing? So their research organization created a mechanism for capturing attention, for capturing memories, and creating a way for us to remember and map short-term memory. It's like a short-term memory system that allowed us to get out of this business of direct mapping. And that became what today we know as a GPT. But in the interim, it actually was a model. There's a, a, a book, or a book, a white paper called Attention Is All You Need. I would look it up if you, if I were you, if you haven't read it. And now of course there's beautiful, you know, Google, Google's NotebookLM. You could just throw that PDF, that white paper into NotebookLM and have two very nice podcast people tell you what it means. But that paper changed everything, right? It changed the framework, the architecture for how we interpreted language, how we sent human language into a machine to be interpreted. That created what was called a transformer, and that transformer architecture is what led us to GPT. So the T in GPT stands for transformer. So what does that actually mean? Well, this transformer has a very deep architecture. We won't— I won't go into all the details for you, but it does create a mechanism for pre— like, for processing human language and then mapping it to a generative model that can guess what it thinks the answer might be. Now, GPT specifically, not all language models are GPTs, by the way, but most of the time when people talk about large language models or LLMs, they're thinking about this GPT architecture. And the first letter of GPT stands for generative, which is super important, right? Generative means you're gonna make stuff up. That's the job of the models to create an answer, not to tell the truth, not to represent humanity, not to be safe, not to protect data. Its job is to generate an answer for you that's pretty good, that's plausible. The prediction of how good that is is the work of the, of the transformer. Now, so GPT, generative. The next letter, P, is pre-trained, and that's an equally interesting part of the framework, which is now we've got a large language model pre-trained on the internet's data. And I don't know if you know a little bit about the internet's data, but it's kind of awful. Some of it's good, some of it's bad, but the internet's data includes opinion pieces. It includes fact pieces, and worse, it includes opinions stated as fact, which is very confusing for an AI model. It's very hard to understand. It doesn't really have any ability to understand truth unless that truth is trained into the model's behavior. And that's why retrieval augmented generation was created. So this large language model, now that we understand the basics, right? Generative, pre-trained on the world's data on the internet. And it's a transformer, meaning it has the ability to make this prediction. When we say something to an LLM, like, I tell me, I don't know, the year that Alan Turing created the Turing test, right? It's gonna go in and it's gonna break up all of those words into tokens. And, and what those tokens are are representations of human language. Because at the end of the day, the machine does not actually understand human language. It understands math. So there has to be a process that takes what we said and turns it into math. And that's what tokens are. And as a matter of fact, the pricing philosophy around LLMs is based on this token framework. So you don't have to know too much about it, but you should know that a token does not equal a word. A word word does not equal a token. A token is a mathematical representation of part of a human, of human language, of part of a word, part of a syllable. Some words are short enough that they represent an entire token, but it's a mathematical algorithmic equation. So it's not something that is usually equally correlated to the language itself. I will give you, I don't know if it's in this deck, but there is some really cool data If you look at all of like 5 or 6 human languages, you know, Spanish, French, Japanese, Mandarin. If you look at these languages, their token counts are very different, and English actually requires the least amount of tokens. I know it's kind of shocking. So you could do a little research on that yourself if you want to. I'm just gonna— oh, I can't see the chat. Was it a question that came in the chat? I don't know. I will look in just a moment. Unless you all see that it's a question. Hi, come on in. What do you need? Oh, well, I am teaching a class on— it's called the Global AI Bootcamp. So we're doing LLMs, artificial intelligence. Yeah, you're welcome to sit and listen if you want. Yeah. And if you want help in like the facility, those people over there can help you. Yeah. Yeah. Yeah. Because this is like, yeah, they do a bunch of stuff. And that the guy in the corner over there, he's like runs the show. Kendra, it's you. I see you, and I approve you. Okay. Awesome. So what you can see here is, you know, we enter in some natural language. It turns into tokens, the mathematical representation of that language. We hand it to a model, and the model has to make a decision on what to predict. So this is a really important process in machine learning. If you are not familiar with it, it's supervised learning, or I should say supervised learning is the best way to characterize it. But when a model makes a decision, it has a confidence interval. It knows how good or bad a word choice is. And so there are a couple of different settings that you can use on any LLM to increase the variety of words that model can choose from to create and generate a positive result. So once that's generated, the output that you get, the answer that says, you know, 1951— I forgot the question I asked, but the, the whatever the answer is to the question that you're asking, build me a marketing, you know, model, create for me a, a business strategy. When you say these types of things and it turns into a pretty good-looking strategy, it's leveraging its training data unless other data is available to actually guess based on all the things it's been trained on, what it thinks it might be, right? So it has nothing to do with truth, nothing to do with, you know, what is right. It has nothing to do with humanity or fairness. It just is trying to guess the very, the most plausible next set of words that would generate an answer you'd be happy with. And so that is kind of like the way that it works. It's important to know this just so you realize, like, it's not magic, that it's actually all math. You don't have to know that math, but you should know that this is a mathematical equation that ends in a confidence number, and that confidence number determines which word is presented to you. It's very, you know, I always talk to people and they're like, oh my gosh, this is amazing. It's so magical. I'm like, it's magical to the user. As a matter of fact, there's a science fiction author. His name is Arthur C. Clarke. And he has a quote that says, you know, that technology AI is akin to magic. And I actually agree with him 100%, but it's magical to the person using it. It's often not magical at all to the scientist, data scientist, data engineer, or applied AI engineer that built it. Like, we're like, no, no, no, we, we made that magic happen. Like, we, we tied those things together. There's an architecture and an algorithm and an equation that made that happen. It's not— nothing happens magically there. You know, a lot of times people talk about sentience And if you go research some of these, they become big press cycles. Oh my gosh, our AI is becoming sentient. But the bigger concern is often perceived sentience, right? That you will see an answer and you're gonna be like, oh my gosh, this machine knows what it's talking about. It has emotion. I'm like, well, or it was trained on emotion and it creates such a powerful predictive answer for you that you think it's real. And that's just as dangerous, by the way, perceived sentience and actual sentience. It, the damage is the same. If I think it's real, it's real. We all can probably, uh, relate to, you know, news, uh, cycles and fake news. Like if you believe something that's not real, it doesn't actually matter if it's not real, if you believe it. So that is why it's super important to make sure that these models are built, uh, in a very specific way. So this just gives you an example, uh, of these tokens. So typically you are passing in a bunch of words. Help me find, you know, I don't know, give me directions to the local pub, or help me write a Facebook post. Right? So you pass those words in, it does that magical transition to a predictable outcome, and one token comes out, one huge potentially context token comes out. So this just gives you a sense of what this looks like. This is— there is a website that you can go and learn more about tokenization.. But the most important thing is just to realize that there's no correlation between the words you use and the tokens that you use. And the reason that that's important to know is that you are billed on token use. And so you need to be careful because you, you can't make the assessment. Now in this case, obviously it looks like 43 characters equates to 10 to 11 tokens. Some people though look at it from a word perspective, not character perspective. So if you look at it from a character perspective, you'll always be optimistic, meaning that your token usage will always be less than— always be less than your character usage because tokens are usually more than one character. However, again, you don't wanna use characters or words to give you a sense of pricing. You're gonna wanna use a tokenization process to help you understand how expensive your model behavior will be. And that's, you can go look at that at the platform. So tokenizer, this website is OpenAI's explanation of tokenization. They also have a token calculator. That'll help you identify how expensive your solution might be, which is kinda nice. So you could see here, this is an example of, like, we need to stop anthropomor— and as soon as I type in anthropomorph, it's going to predict that because of the nature of those tokens, I'm going to want to add— there's actually a word for this in the English language. When you add -ing to something, but I can't remember what it is. Maybe one of you is a literary person, but when it's gonna know that you're gonna wanna add -ing to this. And so it predicts the end of that word based on the tokens, based on the sequence of the tokens coming in, and it's contextual. So this is where that memory comes in, right? So remember Transformers, their job was to create short-term memory. So in order for me to know that it needs to be -izing, I would have to maintain the memory of the sentence. We need to stop anthropomorphize, right? I need to know that whole thing happened. What happened on Alexa, and it still happens to this day because it's not on an LLM yet, is that when you said we, if you talk to Alexa and you say, Alexa, tell me what time it is, it'll tell you the time. And then if you ask it, could you repeat that, it wouldn't know what it just said. It will come as soon as it's done accepting one piece of information, it forgets that it ever did that. And so this model that we're talking about, this transformer model, gives us the ability to remember. And it's in these memories that we can create much better predictions. And that's why it's so powerful. So this gives you an example of kind of what the prediction process— so we need to stop. We need— and as you start typing, it's just like if you go to Google and you type in you know, CEOs in— if you put CEOs in, it's gonna drop down a bunch of words that it's gonna predict, right? It's called text autocomplete, text completion. This is just a fancy text completion model. And, like, extremely more verbose, much more wordy. So you can see here, again, just a representation of the numerical values associated with it because We do want to stress the fact that this is based on numbers. It is a mathematical equation, equation. But its job as— and we like to reinforce this only because so many people over the last few years have gotten a little bit confused about what an LLM does. They actually think the answer it generates is right or correct or has been verified or checked in any way. And it is completely a predictive engine. That's all it does is predict predict the next most likely token, which is just a small piece of whatever comes next. These multi-headed LLMs allow you to do that at scale. But that's its only job, is to predict the, the answer that you're looking for. And that prediction can be wrong. And those wrong predictions are often called hallucinations. So interacting with LLMs, that's what I'm gonna share with you right now. Let's go ahead and take a look. We're going to actually play around with some of these things. Now, when you play around with the model, when you interact with a model, we call it prompt engineering. Some of you probably know that term. But we're going to take a look at what I consider to be the most important way and actually ties right into our AI safety system. But what are, what are the most important things you need to do when you talk to an AI model? And there's lots of different places that you can configure instructions. For a model, but interactions is just how you ask it for what you want. I will encourage you though, that as you ask it for what you want, it is intentional. You do have to think about it. Alright, so here is an example. We want to ask a question and get a text-based answer, right? What's the largest ocean in the world? And it generates an answer similar to what happens on Alexa. We pass in a, a sentence and it gives us a predicted answer. Now, ideally, that answer is not made up. We at Alexa, we ended up seeing lots of these answers come back wrong because the model just guessed. Like, instead of 63 million square miles, like, we— I think one of the ones that we proved was wrong was, um, Kilimanjaro. Like, what is the summit height of K2? And Alexa would make up the summit height. Well, that's a thing. Like, it's a no, no number. Like, you could Google it and you would find it. But the model did not answer with that information. It generated the answer. So we had to do a lot of work to force it. And again, that's the value of using an architecture like retrieval augmentation, right, is that I'm gonna say, when you go to get your answer, I wanna augment, not you. I don't want you to use your training data. I'm gonna tell you to go use the data that has the answers in it. So we have the ability now in our architectures to actually force it to use a dataset as opposed to letting it rely on the internet data it was trained on. I, I don't, I don't know. I would say I almost never use in a production AI solution. I never use the data it was trained on. I always wanna augment it. So Retrieval Augmented Generation has become a bit of a, like, catch-all for any project that you'll do. But now there's even more sophisticated architectures as well. Now, similarly, this is text-to-image. So this uses a slightly different model, but the process is the same, right? I want a close-up picture of a cute raccoon walking through the forest, photorealistic texture and lighting. There you go. So cute. I definitely agree with you. Right? So we pass in a text, it generates an image for us. And I'm gonna show you how to do this inside the tooling. Right? Now I can say, this is one of my favorite things is that when I use a prompt, to generate something, I often want to take that image, give it back to the model, and ask it what prompt I used. So in this case, I can say, like, what is on this image? So now I'm asking it to generate text from the image, right? It shows a cute raccoon. And then from this, I might say, oh great, create me a prompt that does this. So if I see something very fancy online that I like, like a commercial, I will take that commercial that ad, and I will pass it into a model and ask it to build me a prompt for that ad so that I can generate my own version. Then I optimize it. But I'm gonna show you a couple really cool tools now. So let me stop here. And we're, we're gonna start first by playing around with these tools. Let me stop sharing if I can. Oh, good. There's some messages. Hopefully— oh, good. Oh, good. Oh, good, good, good. Let's see. All right, someone answered no. Oh, yeah, I remember. Okay, good, good, good, good. Yeah, that's funny. Chandra was commenting on the buckaroos. Yeah, exactly. Terrible. Okay, okay, okay, we're good. So let me stop sharing if I can find my stop sharing button. Let me go down to the bottom.

Ding.

Let's see if I can find it. I don't see it. Where is it? Oh, there it is. I got, I got the meeting chat. Where? It's gotta be over here somewhere. Zoom. Hmm. Hmm. Alright, let me minimize this. I think I found it. I don't even see you all. Where are you? Come back. Oh, there we go. Hold on. It's gotta be here somewhere. I might have a very dirty desktop. Let's see here. Yes, you are not alone if you have more than 50 windows open at any given time. All right, here we go. 2 messages. By the way, this is the email I use for Microsoft Azure. Oh, good. Thank you, Javier. Awesome. Oh, good. Okay, great. Awesome, Rob. Fantastic. All right. I'm gonna go back to my shared screen. PowerPoint, and it's got to be around here. There we go, I found you. All right, there we go, I found one of it. Any— oh yeah, there we go. No. All right, let's see. Zoom help, meeting information, full screen. No meeting. Keep on top. No, let's see here. Minimize. I could see myself. Hi! What is going on? Alright, so I found the Zoom screen. Now I just need to find my, like, ability to turn off recording. That zoom? Nope. That— let's rewind. Alright, currently mirroring, currently extending. Oh my gosh. Stop extending. There we go. What? That was it. Okay, stop. Alright, there we go. Alright, so now I'm going to— oh yeah, there you go. Sure, you can record. Okay. So we now have a, like, crash course in LLMs. Let's take a look at actually building something. So what I'm gonna do is I'm gonna demonstrate it for you in tools that we already— you could go and have access to right now if you choose to. I'm gonna use ChatGPT. I do have a pro account, so it is what it is. I have a pro account, but I'm also gonna show you Sora, which is a new one. And so if you haven't seen Sora yet, it'll give you a chance to play around with that as well. And then we're gonna take a short break, and then we're going to get everyone who wants access to these credits. Hopefully there won't be more than 30. I think that's as many as I have. So we will get everyone access, and then we will go from there. Okay, so let's take a look. First, I need to open up. See, I did that, put that in full screen mode. All right, I'm gonna try and just share my screen and see what happens. All right, here we go. I don't know what this is, but I'm gonna share it. Oh yeah, cool. Okay, awesome. So let's go to— oh, actually, I have a whole bunch of models. Somewhere around here. Here we go. Can you see my ChatGPT? Yes. Awesome. Alright, cool. Let me hide that, make it a little easier. Okay. So any model, you could do this in any model. So you can go to copilot.com if you choose to. You could go to, which is available to everyone. But I'm gonna start with ChatGPT. You can also go to ChatGPT. They have a free account. So this all works the same way. So I always like, my most common things that I use OpenAI for is to create content because there's only one of me. So I'm gonna say create an Instagram post about the Global AI Bootcamp in Miami. And how much fun we are having. Work with me. Okay, is that what I gotta spell? Oh, I don't know if I have to spell it right, but I should. I'm gonna try it. B. Alright, so text to— Hi, what's up? Great to see you. Yes, yes, we have a whole bunch of people, but because it's the, you know, it's 50 people, but they're here. Jose, our host, is here. But thank you. Yes, yes, we're very good. Okay, so let's see here. There's a fun— oh, an Instagram. What did I say? Oh, an Instagram caption. Oh no, I don't really like that. So I want to say, how about for LinkedIn? So the transformer at work means, like, again, if this was Alexa 10 years ago, I would, I would say, do it for another platform. It'd be like, what do you mean, do what? It would have no context to the fact that I just asked it a question, right? So I go back up here, and I don't really like that Instagram. When I ask it, can you do the same thing for LinkedIn? I don't have to give it any detail about, like, it already knows what it's said, and it uses that here. I don't know, let me see. I don't even really like that that much, but I'm so picky. So what I, I like here is like, would you like to tag people? Would you like suggestions? All right, so I like that. Now I'm gonna give you some tips. So I use a few. I use a bunch of models. These are all a lot of the different models that I use. But for marketing, so we've got here like a fun— I'm actually gonna copy the prompt that I used. It wasn't very long, but I'm still gonna copy it. I usually go to Claude to help me write marketing type content. Claude was created by digital marketers, direct response marketing. It's very focused, it's very tuned on that type of content. But let's just see. So I'm gonna paste it in and see what it does instead. I like it already. What? Living the dream. Like, I like— like, you can see, right? Like, the content is actually different. Nothing beats connecting with brilliant people. That's right, y'all, that's you. It's not brilliant minds, that's you. While in— I love this. See, I don't know, what do you think? I don't know, which one would you pick? I like it. And then of course, which one was this? This was for Instagram. So, oopsie. What about one for LinkedIn? A more professional LinkedIn post. I appreciate that. Yeah. So you— I don't know if you know this, but these models tend to fall into a habit with certain terms. One of them is, I am thrilled. Thrilled. If you see someone post thrilled, maybe, maybe they said that on their own. I personally have never said thrilled, like in my own words, like described, I am thrilled. I've never said that to people. But so that is something oftentimes in my prompts I will say, hey, maybe, I don't know, maybe we don't have to do that. Maybe we never say thrilled again. So there are certain things in my system prompt on, on ChatGPT that I've given as memories that I've told it, never say this, never do that. I tell it, don't tell lies, always reference real data, source your information, and don't use words like thrilled. And there's about 6 or 7 words that I've uncovered that it says consistently. What is one? Moreover. Like I was editing a long piece of text and it used moreover like 17 times in 1600 words, which again, I don't talk that way. Maybe academics talk that way. Maybe that's what an editor would do, but it's— I wanted it to sound like me, so I asked it to take all that out. And that's the benefit, right? If I say, if I don't like this, like I can just be like, I don't like the way you started that. Can you, uh, can you try something else? Be ener— this could kill it— be energetic and enthusiastic like Noelle Russell's enthusiastic. But let's just keep it easy. But not over the top. I like that much better. Energizing days of innovation at Global AI Bootcamp. We might see this later today. Thank you all for working with me to create content. So that is the image part, or the text part. Let's take a look at the image part. I'm a big fan of Sora. I've been having too much fun with this. Actually. Oh my gosh, don't look at this one though. Oh gosh, it's— oh no, that was a good one. I don't know, I'm posting live, but these AI systems are doing crazy things. Like, it's bad. I don't know, should I show you? I probably should show you because it's bad. It's bad news. And it's why you have to be careful. Oh man, Sam. So I'll show you all mine. Mine does not look like that. I'm glad I didn't use that prompt. That's pretty funny. Oh, that's cute. Okay. Alright. So pretty cool stuff. Oh my gosh, what is happening? I'm a little bit afraid to keep scrolling, actually. A little bit worried. Little bit worried. Okay, let me stop. Let me stop. Let me go to my media, things that I have created. So many of you know I launched a book. Actually, it launches tomorrow. So I, I think what I love about our new image to text or text-to-image capability is the simplicity of the prompt. This is known as meta-prompting. So basically when I say something, the model is manufacturing a more sophisticated prompt behind the scenes that I don't have to know about. So in this case, all I had to write was create an image of Mark Zuckerberg reading this book, and then I handed it this image of my book, and then it created that, which is pretty crazy to me. But I have— but wait, there's more. And in all of these, one thing Sam Altman said recently is that they spent a lot of time creating these, like, gallery images. Oh, that's— I know that one. Creating these gallery images, right? So they've spent a lot of time doing this so that you can— oh, look, It's from— anyway, I won't tell— let you all know how I know that. But now I can just grab this prompt, right? And I can edit the prompt. It's very similar in the world of AI that I mentioned, the umbrella, right? Applied AI versus custom AI. Building prompts from scratch takes a lot of time. I don't know if you've played around with these tools, but I remember in the early days of DALL·E, I would be trying to build an image and I could never get it to look like the people that do art for a living because they'd use words and shadows and f-stops and all these like photographic and artistic terms, brushstrokes. I didn't know any of that. I'm just like, give me a picture of blah, blah, blah. But what I could do is I could take an existing prompt and apply it to me, right? So that allowed me to— I actually don't know Where would I have done that? I'm wondering if I did this in my— I have so many prompts, but let me see if I can search for— see if I use this model to generate my Lego Noelle Russell. Right? So, but I could not build this on my own. That's pretty good, but it took me a long time. Because I didn't know that this is a 3D anything. I didn't even know 3D— I should use that term. I did know Lego miniature, but I took this prompt from a prompt repo, and then I just edited it to be me. Another one, another fun one was Funko, and I'll put these prompts in there. So if you want to create your own, you can. You can also grab these from Sora as well. Oh, we got lots of stuff in here. Oh, cool. Yes, awesome. Awesome. What? Oh, I know. Y'all are— oh my gosh, I'm a little bit worried about the showing thing, showing you guys what Sora did. It is awful. So here is that, that prompt if you want to create a Lego version of yourself. And I went through multiple iterations, as you can see, but it took a long time for it to get my hair right. Okay, so let me go back and search again. There was also Funko. See if I can find that one. Oh yeah, this is the one, but it took a while. But so I liked this because I think— let's see if I have— I mean, yeah, I don't know if maybe what I could do is ask it. So this I did a long time ago, last year. So I don't even know if the model will be able to recreate this. But what prompt should I use to recreate this image? So this is similar to that part in the presentation where we said, give it an image, but the image is already here, and ask it to reverse engineer it for you, or you could say what's in it. But I'm gonna try and do something a little bit more fun. But look at the hair, like, that's pretty good. I was so excited. It took me a long time. Uh-oh. That is— oh, I don't even know what model I'm in right now. I might be in DALL-E, so it doesn't know. It can't, it can't actually give me words because it's an image model. It's not using like 4 Omni. Oh, there it is. Oh, That's weird. I think it did it as an image. Alright, let's try this. So I'm gonna move this over here. Ctrl+C. Oopsie. Oh man, it won't let me copy it. What is wrong? Ctrl+C. Oh, there we go. I think I copied it. Alright, let me go into the meeting chat. Oh, thanks, Sherilyn, for posting that. Yes, if you haven't gotten my book already, I did not bring it, but I did bring shirts. I'm just saying, if you're here live, you get a t-shirt today. Um, all right, yes, because I, I missed it last time, but I was the first thing I put, but then I forgot my box of books. So, you know, you get one thing or another. Um, oh yay! All right, cool. Dang it. Um, but at least I got a shirt, but it would have been nice to trade that for a book. Uh, but we're going to be doing more lives, so y'all can come see me. Uh, and I'm also going to be setting up meetups, uh, in every city that I'm traveling to on the book tour. So the day before— so for example, this coming week I'm going to be in Charlotte. So the Thursday— Wednesday night we're doing a meetup at a local Starbucks. We're just going to hang out, have some coffee, um, and get to meet all y'all. If you want— if you're in Charlotte, you can come and visit. Um, okay, here we go. Uh, oh yeah. Oh, I don't know if I— oh, I think I just put that in there. Logo trademarks. I don't know if that works. Is that the right one? Yeah, vinyl figure in the style of Funko. Okay, great. So these are pretty, like I said, you can use these prompts. But I liked it because at the time I did this, it like a bunch of people started doing it. So the whole LinkedIn was like filled with people's Funko Pop thingies. So it was pretty cool. Yay, Evan! Awesome. Oh my gosh, I can't wait for it to hit you too, Evan. You'll have to tell me how you like it. I actually— oh my gosh, I can't believe I didn't even bring one. But I started doing this thing, which I don't know, I guess I'm trying to validate that this book is actually decent and it's got good stuff in it. So I would just randomly open it to a page and just read what's on the page and be like, please make sense. And it does. I've done it like 10 times. So I'm like, this is good. This is good stuff. So I'm pretty proud of myself. I'm very excited for the launch and cannot wait for you all to get it. Yay! Awesome, Kendra. Thank you. Yes, it's exciting. And so I will apologize in advance. My algorithm will be all about the book for the next 30 days. Like, you will get nonstop. So if you've already bought it, I know you already bought it, but it's not— it's just going to send it to all y'all and your friends and like up to 6 networks deep. Just be excited for me, but I apologize if you're like, ah, Noelle, already. A mantra I've been told as a business owner is pretend like no one sees it. And the benefit of thinking that way is that you all have seen it, and you know, thank you for supporting my dream to become an author. But so many people haven't seen it, and the only way for me to get them to see it is to continue publishing. So yeah, yeah, yeah, exactly. So cool. Okay. All right. So you got that idea. Let's go to Sora. So Sora is doing some pretty cool stuff. That's my baby tiger. Some of you might have seen these. The benefit of using these tools from a content perspective is that they're, they're scroll stopping. When people see stuff that's weird, like a baby tiger hopping in the snow, they stop and they read a little bit. Now the next job, now that I've gotten them to stop scrolling, my headline has to be key. Like, it has to be locked in. It has to be something that I say and they're like, what? I have to read more. And that's where I leverage AI models to really help me, um, help me do that. Um, yes, 100%. And yes, and I would love to help other people write more stuff, 100%. There are not enough women in this space. I mean, I went to an event and I was the keynote and I did a book signing and a woman came up to me. She was like in her 60s. She was older than me. And she's like, I've never met a woman, like a technical woman author. I've never met one. I was like, oh my gosh, let's go. So yeah, so if you have it in you, this is your time. OK, so let's go continue on in the content journey of things that I have built with this tool and technology. So this was a prompt that I saw that I liked because it was a, it was actually a puppy sitting in a chair drinking a cup of coffee. None of it was labeled as it is now. But on this, on the screen, it was actually a television showing like an urgent crisis, like a seal, like it was a seal being eaten by a whale, and it was like climate crisis. That's what it said. So I rechanged all of that and But again, this prompt is pretty verbose, right? Like, I don't write all that. I just went in and changed the stuff that I wanted. And here's the fun fact. I changed tiger like the first 2 times. I didn't realize it had tiger. It had the word dog all throughout it, but just changing it the first couple times changed it everywhere. So even though I didn't catch all the, all the options, but yeah, like, the tiger is calmly looking at the small wall-mounted TV. It doesn't really look like a wall-mounted TV anymore. Like I made it into like a shadow box somehow. I don't know with the words I chose. But I wouldn't— it would've taken me forever to build something that looks like this. So now I caught— this is the concept of applied prompting. Take someone else's prompt and then make it your own. Those prompts are not copyrightable, at least not yet. So it's not like you're stealing. That's why the prompts are made freely available in platforms like Sora. So it's pretty cool. All right, so let's go and see some other stuff. So this was me testing out text creation. So one of the things that image models are really bad at is reforming text. How many of you have been burned by like, you go in and you're like, build me a sign for this, or create, you know, it adds text to your image and it misspells it, or all sorts of nonsense happen. And so, yeah, exactly. So that, those days are gone. Like it did pretty good. This was one shot. You know, a one-shot prompt. Very simple. I copied another prompt in order to get there. I don't know if it allows me to see the original.

It doesn't.

But it was, yeah, it was, I was pretty impressed. Mem. And then I have to look at these words. I'm like, is that really spelled right? I can't even tell. But that is that. Let's go down and check out some more. So these were pictures of my kiddos. That's my daughter Jocelyn as a Android Mona Lisa. Pretty fun. Um, and it then, just so you know, I uploaded a single picture of my child and then hit a content flag that said do not upload more pictures of kids, uh, which I love. Thank you for doing that, Sora. Appreciate you. Um, but my daughter was there and she wanted to play anyway, so I let her have one. Uh, and unfortunately I have lots of pictures on Facebook of her, so the algorithm knows her face. Anyway, there's no hope. This was me. I sent in my picture of myself in order to create this. It doesn't look like me at all, really. I don't think so, but it's cool. It's nice wearing red. That's cool. But yeah, it doesn't really look like me. Awesome. Thank you for that. Let's see. Oh yeah, this is where I tried to upload a picture of my daughter and they were like, our content violated. This content violates certain policies. There's my friend Satya reading my book. Thank you, Satya, for all the thank you. Of course, he did not ever read this book. So I posted this on Twitter with a bunch of, like, other pictures that were manufactured. But, like, this is crazy, right, that I could post something and even put it on a story. I don't know, like, I feel like Satya would be like, Don't say I read your book if I didn't read your book. I don't know. I don't know. Seems weird. Same with Mark. I mean, Mark Zuckerberg, I don't know. I feel less, less like I'm hurting him less somehow. But I feel like Satya is like a nice guy and would be like, hey, man, don't, don't lie. So, but it's interesting that it just allowed me to do that. And it's like an advertisement. And like, what are the rights here? Right? Enter responsible AI. Like, what? Are the rights that I can grab any celebrity and have them read my book. Ryan Reynolds is next. We're doing that. Anyway, it's just fascinating to me. So then I wrote this very simple prompt. This again, just like, look how simple it's one line. I could never do this before. It would take me hours to come up with a prompt, or at least search on Google for a good prompt. But these are single-line prompts. It's like one-shot prompting, one-shot, one-shot learning. It's crazy. This is me again. Doesn't look like my face, but this is my version. I didn't get all buffed up like Sam Altman did in his. That was a good idea. I should have thought through that. But this is my Pokémon card. It's cool. Like, create a Pokémon card of yourself. Just try it. The prompt is a little bit longer, so I'll throw that prompt in here. And I gave it a picture of myself, so it should have been better, but that was the original. Image that I took. But I'll throw it in there in case you're interested in the prompts that I'm using. So, so this, I mean, we are so far accelerated from, you know, the first time we started using this 2 years ago. Like the amount of things you can do with no skills whatsoever. Before you had to kind of be an artist to build good artistic content. Now, like, I just have to I mean, a lot of this stuff, like, I don't know what this means. I edited like what I was wearing, but like Pokémon style trading card. Okay, that's cool. I didn't— I don't know how that is. Only part of a thumb visible. That was a good idea. I wouldn't have thought to do that. Let's see. Oh yeah, layout abilities. The card includes fitting Pokémon styles, text layout abilities, stats. Yeah, so I wouldn't have known to do all of that. That. But thankfully someone did, and I'm borrowing from their beautiful— building on the shoulders of giants, as we say. All right, let me close that part. This is, uh, voxel art. I don't know if you're familiar with it. It's a bit trendy. Um, so I just typed— put in my headshot and asked for voxel art, and that's what it did. You could do this for anything. But again, just think about it from a social media content perspective. Like, I can generate content on all these different styles, and it takes me 15 seconds. Like, it's crazy. Um, these were my early attempts before I realized I had to tell it who I was. Um, so it was just manufacturing what— I mean, the shirt was nice, right? I like the outfit. But then I had to like, no, you have to look like me, curly hairs. This is probably my favorite. This is Muppet Noelle. Um, but there's some nonsense that went into this. So I was in the video part. I'll show you a video in a second. Video is crazy. Hallucinating. Mutilations galore. But the image part is pretty good. So this, I'm actually thinking this is gonna be my new Nearly Noelle. We're gonna do kids AI training. I'm using MIT's curriculum and, and this little Muppet person is gonna like, I don't know, we'll see if I get a cease and desist from Sesame Street. Like it could happen. I hope it does. Like that's a good story to tell. Like you want it like, I don't get fired, but like getting attacked by like a brand, that's kind of a good deal. Like you get lots of credit for that. But it's interesting, right? So if you look at the logo, like it turned it into this like happy face. That's the logo of mine. It's actually like people holding hands in different colors. The logo is Inclusion Inspires Innovation. But I thought it was pretty cool how it, how it got to all of this. Yeah, so that Muppet is probably my favorite. Um, oh yeah, so here, let me just show you how this— oh gosh, let me not, um, I'm about to get to the bad stuff. So if you are easily triggered, I just want to warn you, like, some of this is not good. Models do bad things. It's not like X-rated, but it's definitely R-ish. Maybe it's PG-13, but it's definitely not for kids. Um, so this, if you look here, this video like, turn her into a Muppet. That's what the prompt is. This is what it did. First, it took my picture and made it animated. That's pretty cool. But like, what? Like, what? Like, is, is her name Muppet? I don't know. I don't understand. Don't understand. Um, so that's what it started with, and I was like, what is wrong? Now I realize I said turn her into a Muppet, and I'm in video, and I thought it was an image, so it's trying to create a video. But still, How is this person a Muppet? I don't know. I don't know. I don't like it. It makes me uncomfortable. So that happened twice. Then I was like, all right, you didn't get what I was saying, so turn her into a Jim Henson Muppet. Different Black woman. I don't understand. Is it the hair? Like, what is happening? I don't know. I would love your opinions. It's crazy. All right, get ready. It's happening. I'm about to show you this nonsense that I did. So I have— So many of you know I have a bunch of businesses, but one of my businesses is being a marketing, um, company. So I have happened into chiropractors. I don't know how it happened, but I have 3 chiropractors that I run their marketing for, and I use AI to do it in less than an hour and a half a week. Um, and I build all of their socials, all their emails, all their, um, uh, their SMS messages. Like, I create the entire sequence all done in AI. It's all quite beautiful and they love it. So in order to stop the scroll, I was like, let me build some cool video with the cool video capabilities here. Are you ready for this nonsense? All right, this is how it ended, which was not terrible, but this is how it started. Oh my gosh, here we go. I hope I don't get in trouble for this. Here we go. Are you ready? Don't know what's happening. Okay, that's all. I'm only showing it once. Let's go to the next one. Oh wait, that was a good one. This is the next one. So it looks nice, innocent. Anyone like, what is happening? Like, what is happening? Like, do you see that? I don't understand. So then I was like, okay, no, no, like, support, no bed at all. I'm not going to do anything with a bed. And this isn't terrible, but do you see the hallucination that's happening? It's very Inception style. It's actually kind of cool. It would stop a scroll, right? So you do have to kind of think, what is the point, I think I'd probably— I don't know, it's not terrible. I'd probably keep it because it's cool. It's cool and it makes people go, what the heck am I looking at right now? But the other ones, all I said— I have to— I have to go back to like, um, maybe I could pause it. Let's pause it. Oh, oh no, I can't pause it. Um, don't look. Okay. Um, yeah, create a video of happy patients getting care at a chiropractic office. My husband had a perspective on this that I'm not going to share, but apparently that told the model a certain type of behavior was expected. Yes, get what I don't have this in my brain. I don't want this in my brain. It's craziness. So here's the moral of this story. Like, use these things with care. Um, and testing is required, like required. Perfect. All right, so there you are. There you have it. That's like the craziness that will ensue if you are not careful. I will tell you though, I did— oh, let's see, that was the first one. Oh yeah, here's another one. This one was not terrible. It was kind of like, yeah, it looked like happy— I think happy customers checking out after a chiropractic visit. Like, at least they're dressed well, you know, like there's some diversity. Like, it, it's good. This is a one-shot first prompt generated. But geez, with the other stuff. So you all have to be careful with this stuff. It's very dangerous. So the good news is, is that there are some safety processes put in place. They didn't catch that, but there are some safety protocols put in place. So what I'm gonna do now is we're gonna take a quick break, about 15 minutes. It's 12:44 on my clock. We're gonna take 15 minutes. I'm gonna work with the people that said they would like access to Azure. I'm gonna send you— you have to actually set up your own account, but I have a credit code that you're gonna use. So I'm gonna give you that code. You'll— I don't know how many credits it is, but it's enough to do what we're doing today. Okay, great, Kendra. No problem. So I'm gonna take a break. I'm going to maybe get out of this crazy world that we're in right now. Maybe I'll go to, like, Global AI Bootcamp or something. Maybe I'll go back to DALL-E. There. Oh yeah, see, I've like— my anxiety just lifted, like, being on that platform. Um, anyway, so we're gonna take about 15 minutes. I'm gonna set you all up. And, uh, oh great, Char, I see you too. Oh good, Tony. Okay, awesome. So let's take a break. Go take a walk or play around with this technology. Don't try to get it to do anything crazy. It will do crazy stuff all by itself. But definitely use some prompts. And if you want a challenge, try and create a Funko of yourself and then share it back to us. When we come back. All right, so ready? Break. All right, I think I can do a timer. There we go. What? 15:00. There we go.

We are on the clock.

Okay, let me— I know I'm doing the intro to live coding. Can I send them somewhere? Yeah, yeah. So it's actually going to be continuing education. It's going to be every Tuesday, Thursday, um, 2 to 3:45. So Ernesto and I, uh, created a syllabus. I'll teach it. So this is— then it's going to go through the certification process as of now. Oh, go ahead. So, but maybe by fall. No, it's probably going to be the next fiscal year. So we're going to say fiscal year, everything has to be in through by September or August, whatever. You don't think you'll get it in by then? Um, but if you get it by September, that's for the next. Yeah, let's see. Yeah, yeah. Oh, I thought that was the end of that. No, it's 5:30. We're going to be doing Bolt's windsurfing cursor, but it's mostly like design thinking, full lifecycle. So I'll be going in, and I'm not sure if they posted. She went on vacation last week, so we just got everything approved and still have it. Should be, should be posted on all the channels next week. I was saying is we see super fast. Yeah, you see like the AI things in the— yeah, yeah, yeah. And that's the whole reason why I'm doing the live coding thing, to take anybody from off the street and teach them everything they need to know for any type of development. So we're going to start with understanding anatomy, design thinking, why, what is— coding zero. Oh, okay. So, like, really I wanted to replace AI with the headless understand what, and then understand what some of the main tools are, how do you do that. So I created a way for, like, really low bar entry with super high ceiling. What are we going to have to do? Are you— you still get something Okay, friends. Hmm. Not yet. I'm giving them out now. So if you need credits, I'm going to— now again, a few, a bunch of you are in the AI Leadership Institute. Not you, Adam, because I haven't added you yet, but I will. But if you are in the AI Leadership Institute and have gone through live learning labs already, So that's the benefit. One of the benefits of our AI Leadership Institute is as a member, you get access to an existing tenant where you get a subscript, like you basically get a license for Power Platform, which is what we're going to talk about, AI Builder, Power Platform, Copilot Studio. So you get those licenses as part of your membership. But because this is a free boot camp, I have credits to give away. And if we run out, I will give you access to my platform. I'm not gonna let you not do the thing. But hopefully we won't run out, because I did it based on how many people registered on Thursday. So hopefully we don't have a lot of new people. Okay. Alright, awesome. So I'm gonna just go through now, and I'm gonna direct message you the link. Please do not share it yet, because Like I said, you're gonna, um, oh, let me do this. So I'm gonna direct message you the link. It's going to include a code. And let me see, um, credit. Bye! I'm super excited about your vibe coding. There's vibe coding coming to Miami-Dade, everyone. Um, so super awesome. Yeah, please do, please do. Okay, I could promote it like crazy. Um, but yeah, awesome. Are you going, Jose? Oh, sure, for sure. Sounds good, sounds good. Um, all right, some of our people are taking off, but I'm gonna get you all set up here in the next few minutes. So let me— I'm just gonna roll through. Um, so Derek, you were last, so you end up being first. What? It says I'm not allowed to copy it. All right, wait, wait for it. Copy. Let me do something real quick. Notes, paste, and then copy from here. Copy, Ctrl+V. There we go. All right. So, and Derek, if you can be— oh, there's some weirdness in the URL. I'm so sorry. Let me try that again. There's like a— this because I copied it from an image. We were just talking about that. All right, there we go. That should work. So once you are good, use that code. That should give you access to the lab, and then you are all set. Okay, so I need to just keep track of how many I'm giving away here so we don't run out. Okay, so Derek, you're good. Let's see who's next. Javier, remember to mute yourself. Oh yeah, sorry. I thought I did. Oh, it's because someone talked to me after. Anyway, yeah, thank you though. You're right. Um, wassup? Good to see you. Um, okay, sorry, focus. If you have a moment— oh, of course. So much good stuff. Okay. Char, I think you're next. See, there you go. And then, yeah, if you need credits, just throw me a message if you have not done so already. Otherwise, I'm going through right now. Masa, is that right? Is that how I say it? Okay, great. There you go. Oh dear, I copied the wrong Let me recopy this. Ctrl+C. There we go. That should work.

Uh-oh. All right.

Oh, no worries.

Oh, yes.

Sherilyn. So Sherilyn is here. Let me send— actually, Sherilyn is here. Can you— oh, I don't know if you can ping her directly. Share this information with her really quick. Sherilyn Chandra Brown, can you find her message? She needs access to the Bot Builder course. And as a matter of fact, I think Kendra also needs access. So can you grab their information, find them in the school? You did join the school, hopefully. So you have to join the I Love AI school. And once you're in the school, we grant you access to the Bot Builder course. So everyone who buys a book gets access to the free on-demand Bot Builder course. So if you have not gotten access yet, let us know. Okay, where was I? Sorry, that was that. Okay, Masa. Oh yeah, shall you see? Oh my gosh, almost sent it to Masa. There you go. Oh, oh, man, it should be the same code for everybody, but I would love to know when it's working. All right. Oh, good, then we'll find you, Kendra, and just add you. It's just a private class, so only people who bought the book get access to the on-demand version of the class. Oh, thanks, Sherlyn. Sherlyn's following up with everybody. Okay. Oh, that's how you request it, just saying you need it.

All right.

Oh, I didn't realize I needed to join the school. Oh yeah, that's okay, Chandra. Sherlyn will communicate with you about that to get you access to the school. Yay! I just gave you credit. Looks like I have to head out. This will be recorded. Yes. So Linda, hopefully you're still here. That was 2 minutes ago. It will be recorded, and it'll be cut up. So a lot of these blank spaces will be removed, and you will also get access to the labs there as well.

Yay!

All right, I'm just going through all these things right now. How do you request credit? There you go. Justine is here. What? Hello! Oh, my timer's up. Oh good, it— well, that's pretty loud. Thank you. Um, this is the screen that you will see. Let me— I'm almost done giving out credits. All right, it just got to Justine. Let me see. Linda, I'm just scrolling through. Sorry everyone, not the most efficient. Need credits, thanks. I think I just gave you some. And then Derek got Awesome. Javier was telling me nicely to mute. I'm still going through. Oh, there we go. Julia, I think you have an account on AI Leadership Institute, but just in case, this is better. Because you also, I think, have been on calls where it took us an hour to get everyone access.

So hopefully this will be better.

Sure. Toni, did I give it to you yet? I'm gonna send it to you now. Okay, so I have Toni, Julia, Rob. You have them now. Thanks, Rob, for bubbling yourself back up. You were just first, and so that made you last as I was scrolling. So thank you for that. Okay. So yeah, sorry, if you're not first, you're last, or first. Um, okay, so that Rob got yours. Let's see, Thriller. Yeah, I'm thrilled. Um, that's funny. Tony, Julia got it. Um, Andrea, did I send it to you yet? If not, I'm sending it to you again just in case. I think I did send it to you, but it's the same link, so. All right, scrolling. Give it to Justine and Char. And oh good, yay! Oh good, thank you. If you put it in earlier, can— and you have not gotten them yet, can you do it again so I don't have to scroll through all these messages? Just let me know if you need it. I just sent it to you, Hajar. So you should be good to go. I'm gonna wait. I'm gonna keep scrolling up just in case, but—

oh, there we go.

Oh, Kendra, I already got— oh, Hajar, I think you were the last one. Derek. Linda. Did I give Linda access? Oh, Linda left. I think I did give her access. Tony, I think you got access, right? Yeah, I think we got everybody. All right, so we're in. So I don't want to really redeem this, but I'm gonna have to redeem one. Noelle Russell, should I use my AI leadership one? It's up to you, but I can, let me, I can send this to you so you can test it out. Okay. Sky, find you. There you are. There you go. Hi, Noelle Russell. This time here with Kendra. Hi, Kendra. Do you want a code? Yeah, because I want— so I'm not sure if I need to shoot that at their leadership or what should I do? Yeah, yeah. I'm going to send you the code right now. Okay. Oopsie. Oopsie. All right, Noelle, same here. Efoa. Oh yes, of course. Thank you. Um, can you just type something into the chat so that you bubble up and I can just click on you? Okay. Jeffrey, what? Jeffrey and Justine are here. Yay! I wish y'all were here in person. Aren't you all in Florida? Um, absolutely. Let me go back and— did I copy that? All right, thank you for doing that. I'm just copying these in and then we'll get started. Ctrl+C. I hope that worked. Yes. Hello, Noelle. Hi, it's Marian here. Would you like some? Yes, please. Can you just, uh, put a message in the chat so I can click on your name? Thank you. I don't see your name in here. Where are you? Did you send me a message? Oh, there they are. Thank you. Thank you. All right. Oopsie. We're doing pretty good. And there's Marian. I see your message. Hey, Noelle, this is Cynthia. How you doing? Cynthia, we're all set if we're in the Institute, right? Like we did it last time on our bot building. Okay. Finally got access that way. You're good to go. And you can still access it. Yes, I just tested. Great. Cindy, did I send this over to you yet? I can't tell. I'm gonna send it again just in case. Okay. I think I got everybody. Speak now. Well, you don't have to forever hold your peace. Yeah. So good to see you guys. Awesome. All right. I'm just wrapping myself up. Uh, I would be sending the— I've got your invite code. I would be sending the request in like 2 minutes. Yeah, there's no timeline for that, so you can use it whenever you want. Um, I just want to make sure you got it, so no problem. Sounds good, thanks. What? So Noelle, this is Sky. I still didn't get it. Oh, I tried again. Hold on, maybe I didn't hit submit, or maybe I selected someone else's name. Oh, you know what, I think I might have selected bot. Hold on, Control+V. There you go, it says you got it. Do you see it? Let me know, Sky, if you see it. It's in the chat, in the meeting chat. I, I don't— you could direct— yeah, I don't have anything. Maybe it's me. I'll try again. I'm on my phone. You know what, let me, um, text you then. Because we— I know you. I can't text everybody, but I could text you. Um, paste. There you go. Oops, except that link is not good. Sorry, hold on, let me fix it and I'll send it to you again. There you go. There you go. I just, I just requested. Oh, thank you. Oh yes, there you are. I see you. There you go. There you go.

Huh?

Thank you. You're welcome. Did you send me that? So what do we do, Noelle? Do we just go to— I just wanna make sure everyone got access before we move forward, but then I'll show you what to do. No, no, I mean to get access. So do I, I— just click the link. In the chat. I did, and then it took me to a juror, a quick start. Yeah, a page that has— asks you to submit your code. No, it didn't.

Oh, you know why?

Because your link was probably broken. Sorry. So there's like a dot, uh, in that link. Hold on, let me resend it to you. Okay. Amazing. Wait, uh, who was I talking to right before the— oh, that was Adam.

There you go.

And then Marian. It was Marian. Oh no, I know Adam was also talking to me. Sorry. All right, let me try and find— the names are not alphabetical, so I just have to like find you. I found you. There you go. Try that link instead, Marian. Thank you. Yes, please let me know. Yeah, that was, yes, thank you, Char. Yes, I messed up my— that was because I copied it from an image and the image did not have content understanding. So it just threw some period in the middle of my URL. Yes, thank you. Sorry. So if it didn't work, you could look at that or I could resend you the link, but actually, That's a good point. So everyone. So that is the working URL, but if you don't have the code, you won't be able to get credits. So definitely let me know. Okay, so once— am I sharing my screen? I think I am. Yeah. So once you are there, you will— once you put in the code, you should see this screen. And so I'm going to click on it, go to the Exploration Hub. And this basically is like an education version of Azure where you can sign up for Azure for Students. I don't know if it requires you to have an email from an educational facility. I've never done that. That part. But we're going to— I will be saying— go ahead, sorry. It asked me for that. Yeah. Yeah, so you don't need to. You could do everything from here. I'm gonna go to software learning. I'm just gonna try computer vision. Yeah. Noelle, sorry to interrupt. No problem. So I got access But then my Azure was in French, so I went to settings and changed it to English so that I can follow up easily with you. But now I can— I don't know, I don't know anymore how to go back to the lab. Oh, okay. I will— Do you mind walking me? Yes. So let me go back here. Oh, there's a link in the chat. Yeah, this link is not taking me to the lab. The link is taking me to the same screen, but when I use the same code, it's telling me that you are already part of that lab. Okay, so hold on. Let me— oops, go forward here. Let me grab this link and give it to you. Thank you. I mean, I don't know if it's gonna work, so I'll try. Yeah, hopefully because you're already in this environment, it should work. I hope. Let me know. Yes, thank you. Yes, you're welcome. I feel like it still is gonna have to— like, you still have to log in with Azure, right? I feel that is what happened. Okay, so here is what I did. Let me go back.

Back.

Here we go. So you should be here. Everyone is on education. Oh, good, Cynthia. Great. Awesome. So if you're in, yes, go ahead and throw into the chat if you see the screen that I am showing you, the education overview screen. This is basically a place where you can build systems. I don't know how much credits we got. They just said it's free access. All right. Yes, yes, yes. Okay, great. Amazing. Thank you. Okay, cool. And then I'm gonna go back to that email that I got, make sure— I don't know what else it says. So now what we're gonna do is we're gonna choose Explore All, and it should have you sign in somehow with an Azure account, which you might have to create one, but hopefully you have one. And hopefully we don't just end up right where— I mean, this whole point was to avoid people having to create accounts, but I don't know, maybe the credits are only— we'll see, we'll see what happens. So I ended up here, uh, where was the Explorer? Oh, um, that was that link I sent out. Oh wait, it might still be in my chat, but this This is the fun part of labs. That should take you to the Explorer window. Can anyone confirm? Oh, sorry, can you send again? Oh yeah, it take you here. Okay, good. We have confirmation. It is possible. Now it doesn't mean it won't— it will be hard. Like, not everyone will get in. Statistically, like, 5 of you will fail to be able to get this far. But, um, sorry, can you send Again, my laptop quit. I just— yeah, I sent— oh, Rob, do you need that code thing again? I can grab that. Is that what you mean? Or did you already redeem the code? Okay, got it. Hold on one second. Let me go and grab it. I'm still stuck at the education overview. Screen in the portal Azure. Let's— here. Yeah, I was too. I think you just hit Explore All on that first box and it'll take you where Noelle is. Okay, thank you. Yes, thank you. So Explore All, which is kind of down here. When you click on Explore All, then, then you'll have to authenticate into a tenant or creative, create a sign-in. So, I got an error, Noelle, but I'm gonna try, let me try something different. All right, and then I'll, you know, sign out, sign in, clear cookies, you know. All right, we're halfway there, most of us are there. Okay, I see Yes, yes, let's see. Alright, cool. So now that we're here, where the fun happens is inside. I mentioned this to you earlier. Azure AI Foundry is where all of the cool things are and exist. Well, that is great. I recorded the whole break. All right, not now. Okay, let's do this. Okay, all right, we're good. Chart's good. Okay, so now I'm going to click on Azure AI Foundry. This is I'm just gonna explain what you're seeing here. These are the services that we have created. You will have to create a service in order to support the Foundry platform. So I'm gonna walk you through— actually, I wonder, let me see, hold on one second. I'm just gonna see if it's possible for us to do this without going through all of this. Hold that thought. Where is— Mm-hmm. Yes. Okay. So what we're gonna do is we're gonna create a project. I'm gonna encourage you to just use your name so it's easy and everyone has a unique one. But you're gonna go into the Azure AI Foundry and create a project. Create a project. This is gonna hold all of your stuff. If you were gonna create lots of projects, you'd create a hub where all your projects would be. We're gonna start with the project. So it should— uh-oh, that's not right. Oh, good. You guys have a different screen than me. Hold on. You're not already— you don't have an animal. But you already have them. I do. I have. So they could do a free trial. Yeah, you could definitely. Okay, so if you got a screen that asks you to sign up with— how do you get to that part? It's taking me to— yeah, yeah, yeah. So start the free trial. This— you're going to get 200. I hope this whole thing wasn't just to get 200 free credits. That'd be pretty crazy. Um, but yes. If you get to that page, that means you don't have a Microsoft ID of any kind and you have to have a way for them to identify you. So you'll start the free trial, which is on the— I guess I could log out. There it is. What? All right, I found it. I don't know what that means. Do you guys have that in yours? Drop down your list and find out. I'll show you. Like, when I go here, it says Miami Global AI Community. What does that mean? Do you have that? Oh, what? Oh my gosh. Okay, so we're glad you're here. Get started. Okay. All right, just watch along, everyone. I don't know what's happening right now. How do you plan on using Azure? I'm gonna deploy cool stuff. I do not want to start a tour. Alright. So let's see. Huh. Does that mean all of us will be in here? Oh, you got that too. Just skipped it. Okay, cool. Cynthia, did you end up being in this Miami at Global AI Community thing? No, I just got— I'm on that same page. And mine is in my— let me make sure. No, I'm in my AI Leadership Institute thing. Oh no. Yeah, that's good though. I mean, we, we worked hard for that. Yeah, we did. We earned that. We earned that. That's right. Um, all right. Okay. I got some thumbs up though. Let's see. So quick question. After we put in the code, it says your request has been sent to the organizer for approval. Is That normal, or was it supposed to take me to a different page? Interesting, because I did not create that Global AI Miami. I think that's what the code created for me. I see. Oh, good. Thank you. Super helpful. Oh, yeah, that's it. Amazing. Okay, cool. All right, all right, cool, cool. Let me go back. So back on the education screen. Thank you, Thee, it helps. Okay, so let's go back. We might be too far into the situation. I will, the link that I sent in the chat earlier, see if I can find it for you all. The education overview link, I'm gonna copy it again and paste it for everybody. Everyone in the meeting. There you go. So when you click on this, boy, I guess it takes you here, right? Is that right, Sky? Oh no, Justin, Justine, maybe, or was it Jeffrey? I can't remember who sent me this. And see what happens. It should take me to an education overview section. Yeah, this. So remember this page? If everyone could— the link that I just put in there should take you to this page. So let's get everyone on this page first. But mine says different stuff. Oh, maybe because I don't know. Um, that's interesting. Justine, you have labs. I don't have labs. Maybe because I'm the owner. Hold on, I'm gonna sign out and maybe I'll give myself I'm gonna sign into this and see if it's different. Yes, that's why. Okay, so I was signed into like the owner account or whatever. So this should be what you see. Does this compute for you both? The other one that I saw, right? Oh yeah, I'm seeing labs. Thank you, thank you. Yes, yes. Oh, it's pending approval. Sign up like you have lab details. Yeah, this one's like that's research. All right, pending account acceptance. Maybe that's me. Maybe I'm supposed to go in there and accept y'all. All right, hold on, let me go back over there. I see, it's like double-checking so that I don't accidentally tweet out this code and like the world gets access to these platforms. So that's good news, double protection. But now I have to Go ahead and add you in, and this will just give you enough credit so that you can do the 3 labs that we have. And these labs you'll be able to do on your own. So if we run out of time, you'll be able to take the lab guides that I will give you, and you can walk through it. One of the— the first 2 are pretty easy. The second one's slightly more complicated. Uh, all right, let's see. Title: software. See? Yeah, so Visual Studio. I like all of this. Roles: oh. No, what? I don't know what it's telling me. Um, okay, let me go to Explore All. There's got to be some place that— maybe it's in here— that you all are sitting and waiting for me to approve you. But they did not tell me that. They did not give me this information. So I'm gonna go look real quick. Wait for it. When the hell were you all— where— oh no, dang it. All right, oh no, let me see, let me go back to— thank you for that. Okay, so it does say someone sent me a screenshot. Thank you for that, that's super helpful. All right, 9 months, here we go. Yes, I'm seeing— okay, good, you're in. That's perfect. That's what you should be. Same screenshot. Good. Okay, awesome. Perfect. Oh, education overview, labs. Great. Oh, okay. Yeah, so yours says available $10. Yeah, let me go see what mine says again.

Oopsie.

So I am signed in as Miami. I'm gonna go sign in as me since I redeemed one of the codes and make sure, because it says you have $10 to spend, which means you can spin up these services because they're like fractions of a penny. All right, there it is. Okay, that wasn't there before. Maybe it wasn't. I didn't know what I was looking for. But anyway, alright, so we're good. Okay. Alright. So now, based on whoever you are logged in, and so if you do not have an account, that means you have to create an account. That, I think, was the only blocker for some of you, is that you got to a screen that said start a free trial. So you do have to create an account. Microsoft has to know who you are. But then that account gets associated with the key that you use to get in to get these credits. And you should see in this overview with the link is in the chat, you should see that you have $10. I think it might— let me see here. Yeah, like for those— oh, here we go. It does say pending acceptance.

I don't know what that means.

Oh, yes, good call. Look, see, we are like solution architects. Thank you, Derek. I'm gonna go into my inbox and see if I got a message. Welcome to the Global AI Bootcamp. Oh, oh, oh, no, that's what I sent earlier. Did I get anything from Microsoft? I don't think so. Oh yeah, you've been asked. Okay, so you're going to need to go to your inbox that that account is tied to, and it says you've been requested to take ownership of a subscription. Yes. And you need to accept ownership. So you'll want to say yes. Let me open this up in AI Tools. I'm going to say yes. So it's— it'll ask you to grant permissions for the subscription. It's only $10, but still it's better than nothing. It's enough to do these labs. So yes, head on over to your inbox. Thank you, Derek. You're amazing. And good to know. Alright. And then you'll notice— okay, I'm gonna wait for all of you to get to this page because you do need to edit this page. In order to accept it. You want to make sure it's your first name and last name. Noelle. Sorry, Noelle, I'm a bit lost. Where should I click to come to go to this page? Go to your email. Inside your email, there'll be an email from Microsoft Azure. Let me tell you exactly who it's from. Yeah, it says Microsoft Azure, Azure@no-reply.com. @microsoft.com. So in your inbox, there should be an email from Microsoft Azure. It would have come like around 1:07 maybe when we all clicked that link, when we added that code. Assuming they're only collecting info related to this exercise and not normal work functions. Well, this subscription is complete— it's a separate tenant that you're getting added to. Um, so good question, uh, but it is a separate tenant, so you're not intermingling mingling data. It's a total separate, um, data box. Thank you, Justine. Yes, email, search for the subject, you've been asked to accept. Yay! Thank you. Oh my gosh, you're amazing. So next month when we do this What do you mean, Sherry? Did you— did I send you the, um, I don't know if— did I send you the codes? I don't think I did. Do you want me to send it to you? Are you here to play? If you are, I will send it to you. Oh, okay, okay. Just wanna make sure you're not stuck. Alright. So let's try this again. When you get to the page that looks like the one I am sharing, please throw a thumbs up in the chat. That means you went to your email and did the wonderful instructions that Justine put in for us on this page. Yes, that's true. You will— you potentially will need to— I didn't— I was already logged in, so I didn't have to do any MFA, multi-factor authentication, but you you might have to do that. But it's— yeah, it started here, and then it would say like pending acceptance, and then we went into our, um, into our inbox to accept the sponsorship, which is the credits that they've given us. You'll change the name to your name. And then hit Review and Accept. No, Javier, try again. Refresh. Yeah, I got the same error. I'm trying to hit same. It was something. Yay! Go ahead, Sky. Sorry, I was just— Mine says session expired. Error code, code error message, code 0, uh, something weird. I haven't seen that one before. Oh, sorry. Um, can you take a screenshot and just send it to me? Yep. Um, yeah, so in your— did you go to your inbox? Except that, that'll take you to a different page. Should I click on— should I click on Go to Subscriptions after I hit Accept, or, uh, no, we're gonna go to AI Foundry after that. This is gonna sound like a very rudimentary question, but I don't know how to access my AI Leadership Institute email because I guess that's where it would go to. It wouldn't go to my regular email. So I went to my account, but is it like— a good question because you don't have an email, right? I just use it as a login. No. Oh, sign-in field. There we go. So even after subscription, it says I have to do an Azure free trial. Is that— should I still have to do it? Yes, just the free trial. And you won't put it— there's no credit card or anything. It's not going to send you a surprise bill. It's just to validate your identity. Sign in failed. We had to try again. And I presume you tried again, Javier. Sh— oh yeah, oh my gosh, this is the same thing that happened before. Azure's so hard. So this is, um, so basically what you're seeing, Shar, is they want you to set up an authenticator, multi-factor authentication for your account. So you can put in a number and then it's going to send you a code. This is all security to make sure crazy people aren't going in here and like using AI services to rule the world. So yeah, so just type in your phone number. Noelle, Sam here is telling me to scan the QR code here for the authentication. Yes, just scan with your phone. But what I'm trying to, yeah, I'm trying to scan it with a normal scanner, but it doesn't go through. So you have to open up Microsoft Authenticator or an authenticator app like Google Authenticator. I have the authenticator, but it's just giving me code. Like, so when you— what you'll want to do is on the authenticator, you're going to hit the plus sign. The plus sign. Okay. Yeah. And when you hit the plus sign, it's going to ask you what it is. And so it'll be like a Microsoft— it'll be other, I think. And then it'll ask you for the QR code to scan. Yeah, okay. The plus sign. There really is no way around this then. I don't see the plus sign here. Oh, go ahead, back. It's right, yeah, I'll try and show you. See that plus sign? Yes, but on my phone, the app I have doesn't have the plus sign. Are you using the Authenticator app? Microsoft Authenticator? Yes. Yeah, I'm using— might be at the bottom, or maybe you're in too deep and you want to hit the, um, like hit the three dots that are— or three— the hamburger they call it, like the little menu option. And when you click on that, connect account. Yeah, you want to— yes, I think that's right. Oh my gosh, what is happening? Yeah, because you are attached to AI Leadership Institute. Yes. I mean, they're all, they're all AI Leadership Lab projects. Yeah. All right, so that looks good. Yes, probably at the bottom. Yeah. Oh yeah, if you're on Google Authenticator, it's probably at the bottom. Yeah, you could be in Live Learning Labs. That's fine, because you get access to all— you get more access to this. Your subscription is technically indefinite, as opposed to having a $10 minimum. But don't spend it all. Okay, cool. That looks great. That's where you are. That's where I was too. Okay. So, Noelle, sorry, when I click on the plus sign, I found it. So it's saying, what kind of account are you adding? It's like I need to add an account. I can't see where I need to scan a code. So if you're in the authenticator, when you add a new account, I don't know if I have Google Authenticator. I don't, but let me go into Microsoft. Because I have already the AI leadership here. So I don't have to add anything, right? Because the AI leadership is already listed here. No, but if you got the— I mean, it sounds to me like you're adding a new authenticator. It's asking you to add a new one. Yeah, so Noelle, for some reason, because it's happening to me too, so even though like we set up the authenticator prior with our AI Leadership Institute, going through this acceptance process is prompting us to set up a whole new one. Yeah, so what should I do? Just create, create the new one if it allows you to create the new one. And there's also this option down here that's like this this like QR code button. When you press it, it'll take you to like scanning the QR code. So, actually, QR code button, I don't see. All right, so now I'm going to go to No, I actually want to go to The Foundry. All right, so I'm gonna have a few of you are testing this out. I'm gonna have you go to this link for me. Got it. We're so close. All right, here we go. But yeah, so I did all this to avoid what we do in our live learning labs every single time, and it's the same. So I'm just gonna go back to the old way we were doing it. Okay, that is a link to take you to the Azure AI Foundry page. Can, let's see, I think it was Char maybe and Tony. Maybe Tony, you can test it out for me. Okay, try that.

Okay, good.

I think you're in too. I'm gonna hold the, hold the high hopes that you are also in, Javier. Take me to choose an account page. Oh yeah, yeah, that's right. You're in the right place. So now click on that link, the Azure AI Foundry link. And it should automatically log you into Foundry with the credentials that you have established. And when you are in Foundry, throw a thumbs up. Foundry, but I'm seeing a different set of— You would, because it's attached to your account now. Yeah. Yep. What does it say? Yeah, yeah, yeah. So those were from the last— I mean, you're attached to the AI Leadership Lab, right? Yeah. So those are all everyone else's projects.

Yeah.

So I'm gonna switch this directory to AI Leadership Institute, and those of you who are still in here, Some of you will see this. So what I'm gonna encourage you to do is go up to your name, switch directory, and I have— go see if you can see if you have Stitching Global AI Community. I don't know what that means, but I'm gonna try it. Noelle, I'm stuck a bit again. They're telling me that Uh, the account I'm using does not exist in the 4542. Yes, and so that's what I just showed you. So you want to go up here. I've already done it. You switch directory? No, no, not this one. Okay, let me try. Yeah, so you go up to your name and then choose switch directory and then pick the Global AI Community. Okay, I'm trying. Yes, which directory? Yes, exactly. Thank you, everyone. This is what— and this is like the magic that is Microsoft, is that there's always these directories that you have, like, have to manage. And the more projects you do and the more companies that you do it with, like, these— it's crazy. I have lots of them.

So Okay, I'm at the Azure Services page.

All right, so now click the link. Well, let me put it back in here. Go to AI Foundry, which is— I think it's this link. So my hope today is to equip you all. You will have a platform, and then I'm gonna give you a bunch of labs, and you're gonna go out into the world and do your labs. I feel like I'm gonna be like that, that meme of a cat on the computer, like, yeah. Yeah.

Me too.

All right, we switched directors. I need, um, all right. Sorry, I'm sister. Sorry. That's okay. Who is this speaking? Sorry. It's Sandrine. Oh, hi Sandrine. Yeah, I'm sister because, um, I'm still on the authenticator page. Uh, since I was trying to scan, but it was telling me that it failed. So I tried to click on Can Scan Image. It gave me two codes, I think, one code and one URL. Hi, this is Sherry Lejeune. I called on Friday. Tim never called me back. It's quite urgent that I reach him, by the way.

Okay, I got you.

Okay, you may continue. Sorry. Go ahead, honey. Yeah, go ahead. So I click on Can Scan Image, and now it's giving me two options, a code option to copy and the URL to copy. So I click on the URL, I open a new page, but it told me that it's not possible, or I'm not able to access. Yeah, so if you can click the link that I just put in the chat, that's only because that was my fault because the link I sent you was embedded with my stuff. There was some metadata in there that was my Fortify 42 is one of my projects. So if you saw Fortify 42 is because the link I sent you was, um, polluted. So the link I just sent you is actually from the right place. I gotta do this. Yeah, try this one. Sorry everyone. Got it, Noelle.

Working for me.

8, 7 down, 42 more to go. Uh, so it says you need an active subscription to continue. So should I put the Global Air Community or the Air Leadership? When you go, when you switch directories, you're going to want to switch to This one. This is what I put. I got an error with that link as well. Maybe I should just send you to the Foundry directly. Foundry. This is, this is lame. Yeah, I'm missing from the, this, the global AI community.

Like, I, Tried to switch my tenant.

It's not there. I think I just used the regular link today. I didn't like go through the registered link, so, and my email didn't show up. Hold on. Oopsie. ai.azure.com. That's the easy one. HTTPS. ai.azure.com. So when you, Um, let's see here. Yeah, I'm trying to think. When you submitted the code, did it ask you for some— like, how would it know to email you? Oh, um, okay, there. So there, I sent you a clean link to Foundry. Should be clean. Okay, try that. And then when you get in there, it'll either ask you to authenticate which you'd want to authenticate, or if you're already authenticated, you want it to say— see how I hovered over? You can see it says directory stitching. I don't even know, I think they misspelled it, but globalai. That is the directory you're going to want to use. So you might— you want it— it will be quite liberating when it works though. I promise it's worth the effort. Once you get it done, you're like, I'm a rock star and I haven't even built anything yet. It's very— but it's crazy, this is the same friction we have all the time. Okay, yeah, so now— was that you, Derek, that was talking? I think it was. Yeah, so now Yes, so you were in the Foundry service. We now want to go back out to the Foundry platform. Thank you everyone for sending screenshots out, made it so much easier to track where you all were. All right, so we got a few people in. Let's see here. Current Directory, that looks amazing, Shar. Good job. Well done. All right. We're in. Yeah, you're— so, Derek, I guess you just— you saw the link in the chat, so that's ai.azure.com and you're in. Tony's in. Yay! I know, we're like, finally, sheesh! Cynthia's in. Awesome. And Julia's in. Hey, right? I hope so. If I say you're in and you're not, Let me know. Yeah, that link worked. That link worked. So I was in there. So I signed up for Azure account, another one, right? Yes. Okay. Yeah, it wasn't—

phone—

because you're grabbing a new subscription. Yeah, it's asking me the same. Do you need to elect a subscription to continue? So that— but make sure that if it's asking you that, that you're in the right directory. And if you are in the right directory, that means you didn't accept the subscription. That you got. Yeah, because I don't have an email for that, but now I'm there and it just won't take my phone number for some reason. Says we're unable to validate it. So I'll keep trying, but that's where I am, just so you know. All right, moving right along. Yeah, good one, Blatch. I'll look the email— I'll look for the email in my— yes. Yes, it should say, Justine put it in there, but it should say something like you've been asked to accept or something like that. Let me go in mine real quick. Yeah, you've been asked to accept. So if you want to search for you've been asked to accept, you should find it. Yeah, so when basically the process was you get your credits, then you hit submit, and then the next page after getting your credits was login screen, and whatever you log in, you used was where this accept ownership email went. Let me see if I got a bunch from all of y'all, because maybe that happened. Let me just see if Azure sent me any more. They did not. Same. You have a request to take own— yes, that's it! Accept the ownership. That's it.

Say yes.

Yes, you're in. Beautiful. Okay, we're in. Some of us are in. All right, and then we'll take another break right after this section where we get to play around and do a lab exercise. Let's, let's hope the models are what we think. Okay, so I'm gonna walk you through the first section. First step, and then I'm gonna let you go on your own and ask questions if you need them. Let me go back to my shared screen. Okay, so this is good. Move this, cancel. Okay, so we are here. Mine's dark, of course, because I use dark mode for everything. So yours probably looks a little bit different, but we're gonna hit Explore Models. And inside of Azure AI Foundry, there's what's called the playground. So you have the ability to use models without having to do anything fancy. Uh, you don't have to go through a deployment or build a CodePipeline or anything like that. Uh, that's all been done for you in order to, um, play around with it. So the first one they are going to ask us to use is an OpenAI model. Um, so we can actually— oh, I think it's pretty close to the top. I mean, you can pick one. There's GPT-4.5 Preview. That sounds That's exciting. We could use that. But I think they actually wanted us to use 4.0. So there's 4.0 over here, chat completion. And I'm gonna give you a PDF so you can step through all this. I'm just showing it to you to make one, to make sure these steps work for me, which means there's a high chance they will work for you. Alright. So I'm gonna deploy my model. Select an ex— I don't have any projects, so it's gonna ask you to create a new project. It's just gonna create a project name. It includes your name. I would just keep that and say create. And then it will do all this wonderful stuff. So AI project, I'm good. It's creating the resources to support this. So now what I'm gonna do is go grab that lab guide for you all. I think this is it. See that? Maybe it's in presentations. No, workshop. I guess that's it. I think it's, I think it's one of these. I think it's in here.

I'm gonna go to workshop.

Workshop content. I don't follow all the rules. Technology is live online. This is it. I see it. I see it. Okay, so I'm gonna send you this link when someone needs to validate that you all can see it. Someone please click on it and tell me if you need authentication. If you do, I will be printing this in a PDF and sending them to you. Yes, that's what I'm talking about. Ah, finally, one moment of non-friction. Let's do this. Ride the wave, everyone. Alright, so you should be able to click on that. That is the— we're gonna be doing this lab here. Yay, amazing, Char. Alright, so go have some fun. So what it's gonna ask you to do, basically, you'll see the lab outline. Let's make sure you can get to them.

These are not links.

Oh, yeah. So way down at the bottom, it says move to setup to set up your working environment, which we already did. And let's see. Oh, my gosh! Of course it's gonna use a different platform. Oh, no! Alright. So I think we're gonna just follow the steps. That are in this lab exercise. You will— in this lab, they're gonna also want you to create 2 new accounts. I mean, I'm figuring, like, they said it was free, but I feel like they're setting you up to spend some money. So you're gonna wanna create— do you— how many of you have a GitHub account? I hope— oh, good. Alright. Alright. Good, good, good, good. Alright. Awesome. So, yeah, right, Char? Like, I feel like this is, like, all planned and seeds for you to burn a bunch of money on cloud services. But, but sign up for GitHub. This is good for you no matter what, because all of the models that I use, I put on GitHub, and you should do the same, or at least use it as a place to go search for models. So the lab guide, when you go to set up, it's gonna set up GitHub accounts and a Copilot account. Again, you should have one. You can sign up for free. I'm just gonna go there and show you. You. It will automatically log you in, um, based on— God, it's a different account. Um, uh, so it'll log you in with another— with your account, or it'll ask you to sign up, in which case that's fine too. I'm going to send a notification to myself, which will come to my authenticator. It's nonsense. They don't make it easy. Nope, you don't have to deploy and create a project. You can just start with the link. And so now that you've got that set up, that's one environment set up. Now you need 2 more setups that you're going to do. Yeah. Oh, no worries. Oh, I see. Oh my gosh. So, yeah, for those of you, I need to get back in. So, Jeffrey, do you need like a new code? Maybe do I need to send you a new code so you can do it with a different email? I don't know. Let me know. I'm happy to try that. Okay. Oh, no subscriptions after switching accounts. Okay. So Cynthia, do you need a new code? We have a couple Uh, I'm not sure. It showed that I had a subscription pending, but then when I went through all the things, now it says there's no subscription. Like, if I go back, it did say— um, what are you signed in as? AI Leadership Institute. Yes. So the challenge with that is that you can't get an email. Yeah. And so there's no way for you to All right, so can you hopefully without too much trouble create one off of like a Gmail account or something? Yeah, yeah, yeah. I thought I needed to do the Leadership Institute for the— Oh no, no, no, it was, uh, it's just you had it, so I thought it would be easier, but it turns out you could just— you know what you could do is you, if you're signed in— okay, I got this now. I'm gonna actually go to— I'm gonna add a subscription to your existing account. So don't do anything. Stay where you are. Okay.

And write free.

You, um, yeah, can you help me with that too?

Because for me also, all everyone who you all have AI Leadership accounts? Yes. Okay. Yes, I could do that. Uh, and because I couldn't see that other directory. Oops, that's not the right one. Well, maybe it'll take me to the right place. Oh yeah. Oh my gosh, what did I do? I have so many windows open. Okay, so I'm gonna start with the first one. Let's go here to M365 login. And so for those of you who are working, you want to go to that lab that I showed you. The link is in the chat, and you're gonna step through it. First 2 steps of the setup is to set up a GitHub account and set up a— what's the other one? A— oh, a Chat— Copilot account. So go through those steps, please. I don't want I don't want Copilot. I want this one. Oh no, I don't. So I'm— there we go. I'm gonna find you all now, going to users, active users. Okay, so Who was that? Will it break if you have a prior GitHub account? No, but you can use the one you have before. There's no harm in that if you already have one. I already had a previous account and it worked fine. I started a revolution. Yes, we are all doing it. Okay, let's see here. It's just in. Okay. Okay. Let's see, was it you, Javier? Are you on my list? I don't think so. Um, all right, someone else tell me who is that. It wasn't Cynthia. Oh, was it Cynthia? Was that you? Who was— and yeah, you're the AI Leadership Institute one, right? Yep. Okay, great. Cynthia. There you are. I see you. But it does say you already have a subscription. So, uh, pending acceptance, but I can't, you know, I don't have an email there. Um, yeah, it's Michelle Ryan also. Oh, Michelle Ryan. Actually, you both are going to want to go to the directory AI Leadership Institute, not the switching one. Or stitching or whatever, you know what I'm saying? Let me get back and show you. So, oh, now my AI Leadership one is gone. No, it has to be there because you're— that's the one you're signed into, I think. Yeah, just when I go switch directory. Oh, I can't— I guess I can't switch to the— if I'm in it. Oh, so if I go to switching Globe— Stitching Global. It's loading. Yeah, so you want to be actually in AI Leadership Institute. Okay, I'll show you. Yeah, you want to be in this one if you have an AI Leadership account because you're not going to use the, the, um, the free credits, which now everyone's going to just use my account from now on because I thought using credits would be easier. It turned out to be exactly the same problem. Like, I'm trying though, y'all know I'm like trying to help y'all out, but same exact problem. Um, yeah, okay, because now it thinks that Yeah. And so now if you are, if you're logged in to AI Leadership Institute, now you just go direct to the Foundry because you should have licenses for the Foundry. Okay. And the Foundry was the link you put in here. It's like ai.azure. I don't know. What was it? Portal.azure. No, not that one. It was like, it started with AI. Where was it? I'll look it up.

Here it is.

ai.azure. All right, so I'm going there. So this is where it's asking me to create a subscription. So I just do that with a regular Gmail or my other email? Yes. Hold on. For the Foundry. Yeah. So now when you go up in the corner, it doesn't say you're logged in. Looks like you don't have any Azure subscriptions for the directory for AI Leadership Institute. Perfect. Hold that thought. Stay right there. Okay. It says find an active subscription. I'll be assigning you to my subscription right now. Portal.com. Oopsie. And this I will be doing for the other 2 of you as well, as soon as they get in there. Azure portal. And eventually this will be everyone. So now when people become members of the AI Leadership Institute, we're just onboarding them directly into this, to my tenant. So we don't have to do this. But that's what happens. I tried to— they told me it would be easy. They lied to me. Let's find them. All right, I'm in the Azure the portal. I think I can do this with you all watching because this is important information. This is like the behind the scenes of every applied— like how it actually works. I'm going to take you to the Azure portal. I don't think there's anything that I'm going to show you that you wouldn't need to know. I mean, that would violate anything. So now I'm going to log into my account. She just let me in. I finally got in, Noelle, but I had to do it from my phone for some reason. The Microsoft Edge and Arc did not like the authentication piece for Stitch. Oh my goodness. All right, so I got a subscription. Let's go back here. Home. So users. Look at all y'all. Okay, so let's see. Was that Cynthia first? Oh no, it was— yeah, I think it was Cynthia. Yeah, it was me. Oh yeah, there you are. I see you. Yep, Cynthia, AI leadership. Yeah. Perfect. All right, so now I need to assign you a role. This is what builders do. Yes, like, welcome to the world. It's not all click a button and get an image created. Um, all right, let's see here. Let me go back to the overview and see if I can assign. This is the real work. Yeah, and this is where people like lose their minds a little bit. Um, I don't mind it. I find it enjoyable, but lots of people are like, I can't take this anymore. But Justine has started a Google Doc to, to follow, to document the steps. So I think that's pretty good. That's amazing. And actually, if you share that with me, we can redo this in the community and we can redo it with like in a, in a less frictionful environment. Um, okay, so I got that. Who else was on my list? I think it was Sherri. And then— oh, I have you in here twice. Oh, yeah, the second one. I'm just going to do it for both of you. Who else was it that was on an AI Leadership Institute tenant that needed access? Anyone remember? Maybe not Marian. Marian, you're in on your other one. Tell me. I have two new— oh, it was Michelle Ryan. It's me. Yeah, I keep trying to unmute. Sorry. Um, and you're using AI Leadership, so I got you there. Okay, cool. Yeah, I, I see— I, I'm on a page. It says jump into a project in Azure AI Foundry, view all projects, create— is this where I'm supposed to be? Yeah, you can click on that as long as you were— if you're able to do it, that means you've already got access to— so if I click the plus create project— oh, there's me, Michelle Hub. Okay, so yeah, you might be good then. All right, so I'm gonna not mess you up by adding more to you. So now I'm going to go to not subscriptions, but oh yeah, it is subscriptions, and then it should be here, and then I am— okay, and now I'm going to add the role assignment. To choose— oh, hey, contributor. Oh, sorry, there we go. And then now I can find you. Select members. I do normally when we're not doing this live, I put you all in a group and I add the group to the— like, that's how you do it for real. But because I didn't have a chance to do that ahead of time, I'm just going to add you one by one. So Michelle first. Oopsie. I think that's you, Michelle. Feel it. Maybe I should add both of you just in case. And then, um, who's the other person? I can only keep one, one word, one name in my mind at any given time. Not yet. Uh, let's see. Yes. Oh yes. Um, but you're not in it at all yet, are you? I don't think so. Let me see. Adam? No, you're not in there. Okay, who was it? Was Michelle and somebody else?

Who was it?

Yes, because we were messing around with names. Um, all right, it wasn't— all right, you guys speak up or forever hold your peace. Uh, I'm in school. There was someone in here. Who was that? Javier, maybe? Let's see. I'm just gonna type in names now. Nope, Javier's not in. Oh yeah, that's right, you said you're not in the Leadership Institute yet. Um, okay, crew, thank you so much. Justine's awesome. Okay, there's one more name. Oh my gosh, I should have wrote it down. Cynthia, maybe? Oh, you're in though. Oh yeah, it was Cynthia. Cynthia, sorry. There you are. All right. So how about me? Oh yes, how about me? Uh, Noelle Russell. Who's me? Sorry. Hajar. H-A-J-A-R. I don't think I have you on my list because you're not, um, no, but, uh, because this is like for people who are in the institute where I, like, I created. Oh, okay, okay. Yeah, but if you were, you able to get in the other way, hopefully. Yes, yes, I was able. These people are already in. Name@ailleadershipinstitute.com email, so that's why it was different. Okay. Yes. Thank you. Of course. All right, reviewing and assigning. Please refresh your screens shortly. Dun dun dun. You should be in. All right. Michelle and Cynthia, please try again. You might have to sign out and sign back in. It's not security-based, but thank you, Chandra. Yeah, none of the— I was talking through that before. Nothing I'm showing here is like secured. No passwords, nothing like that. They're all my emails. So yeah, but thank you. I was cognizant of that. But yeah, no, this is fine. This is— I'm just trying to demonstrate the steps that are necessary. But this is a test environment, so this is not a production environment. This is a student environment. Everyone in here is a student. So yes, yes, don't worry about that. Thank you for that, though. Yeah, there might have been, but they're all students, hopefully. If I did, I'll go scrub it afterwards. But usually it's like a they're masked by my plat— like, by the Azure platform. So it's like on microsoft.something. Any chance that the Foundry code can be sent to me again? Yes. All right, so let's go back. Foundry code. All right, so you would be on this thing. I have to just scroll up in our chat and find it again. All the way up here. Yeah, I can tell where I'm at because of the huge prompts that I pasted in. Awesome. So now I see JusticeGPT. Yay! All right, here you go. So should I see something else though? Uh, oh, that was Javier. Uh, I sent it— oh no, um, I don't think so. Let me show you and let me go back there right now. Okay, I just see Justice. All right, so let me go back, cancel. So now you're in the Founder, yeah? Yes, I'm in there. And then I went— I just refreshed and it took me to Projects and I see JusticeGPT and that's it. So, and then let me see here. If you— so you are in the AI Leadership one instead of like when you go to the directory. I can do it now too. AI Leadership Institute. Yeah. Yes. So I should switch to the other or stay here? Mm-hmm. Stay here. And you're going to create the new project from here. And when you create it, you can just use the name, but you're in. You're in. I feel accomplished, but you did all the work. I know, right? I feel like Microsoft gave you a whole layer of extra work to do. 100%. That was unnecessary. Oh, yes. That's right. But that's right. They give me a award for that. So I do it happily, and a sweatshirt. So when they give— you have to make sure you earn that. That's right, stickers and sweatshirt. And I was like, all day, I'll do it. And I love it. But just so you know, and thank you for a couple of you mentioning it, like, it is all for the security of the platform, so I get why. But to learn it, it makes it unnecessarily difficult. These are the barriers to entry. Right? Like, what if we were on our own? This is why I said, like, once you figure this out, go teach it to people. Like, get them to, like, use it. Because if I was by myself trying to do this, you would just quit. Like, that's the reality. People just stop learning because they're like, it's hard, I can't, I'm not gonna— like, especially when it's one of those problems that you, you try to fix and nothing works, and then you do the exact same thing a day later and it works. Like, it's mind frustrating to your brain because you're like, "But why didn't it work yesterday?" And it's like a cookie caching, like all these things start building up and make it very difficult. But once you're in, like I said, it's very liberating. So hold on to it. Don't log out of your system. But thank you for all of that. But like, yeah, so just so you know, the tenant that you did see is a test tenant. It's not production tenant. There is no information. There are some emails in there, so I will definitely, uh, go in and scrub that in the recording. So thank you for that awareness. Um, I did think about ahead of time, I'm like, any— anything. But anyway, um, so thank you for that. Uh, okay, so now we're in. So now you have access, hopefully, to— I gave you all the— a couple— I sent it to you, Shar, by accident, so I don't think you need that code again. Um, but I did send it to you, Javier, so you should be to start over. Any chance the Foundry could— oh yeah, yeah, got it. Yeah, thank you for that, Chandra. Amazing. All right, and then I want to click on this link that you started. Oh, maybe Justine, you can send it to me, your document. So awesome. Um, we've gone through this like, I don't know, 4 times, and every time we try it a little different to reduce friction, and we end up in the exact same spot. So it's a Microsoft thing, because like we are definitely created variants on our approach and it's always the same outcome. But, but we're in, that's the most important thing. All right, so let's go and take a look. There are 3 different things I'm gonna encourage you to do as part of your lab exercise. Now, I, I like, like labs, like this is how I learned to code is I printed out lab guides and went through them, check, check, check, check. And so, So that's, that's the way I roll. But you have a PDF you can print, or you have a, a multi-page Word website that's gonna take you through the exact same thing. It's 3 parts. Part 0 is the Foundry. And the reason I show you this is because this is where all the things happen. Everything you do in the Lab is based on what, what, on the services that are running in the Foundry. I personally think it's good to start in the Foundry. That's why we started here. But when you go to the labs, you're gonna be using what I would call SaaS products that leverage Foundry models. So the first one, like GitHub, you're gonna go into GitHub. GitHub is going to let you use models, but you don't have to know where they are, how they were configured. You don't have any control over the parameters. You have no control over precision or accuracy. Like, you're just clicking a magic button. But Foundry is where you learn how the models work. Models are actually built, where they exist. You can create your own, and you have some credits to play with. So that's why I wanted to show it to you. Most of you who got to this page, I'm going to encourage you, regardless of what your list looks like— it might not look like mine— we're going to create a new project. I already created one. Uh, oh, I don't know if I did in this one, so I'll create another one here. Just whatever the project name is, if it has your name in it, that's perfect. Um, you all will be in the bootcamp one. You have $10 Again, from a token perspective, that's plenty to do the work that we're gonna do in this lab. When you create a new project and you say create, it's now going to build all the resources necessary to support a project, which means creating you a resource group and creating you a space within the tenant in order for you to do this work. And you'll see it show up here. Now, this project is going to have endpoints. So let's say you create a model and you want that model to solve customer service problems. Now you can name this model. Mine's Noelle 5342. And I can use the endpoint that you see on this screen to access that model from a third-party application. This could be from Zapier. If you're familiar with Zapier, it could be from Zapier. It could be from another web application. It could be from your CRM, right? So if you wanna be able to integrate whatever OpenAI service service you are going to create inside Foundry, you're gonna need this endpoint. That's your key to using it in another system. We're not gonna do that. We're not gonna provide that integration, but it's really important for you to see. Okay, so let me move some of this stuff around so that I can navigate. Oh gosh, let me— there we go. Make it a little bit bigger. So just to give you a lay of the land, now that you are in Foundry, you'll see a model catalog. This is something that most people don't realize, that yes, of course you can use Azure OpenAI Service. That's what we're in right now. The Azure OpenAI Service is an N-1 model off of OpenAI, meaning that OpenAI launches a research model and then the team at Microsoft takes that research model, hardens it for enterprise use in production, and then pushes it out on the Azure platform. The benefit of the Azure platform is you get all the security, privacy, and capabilities, RBAC, all the things that we just went through to log in, like that protects the system that you're building. OpenAI, you don't own that, right? That's owned by OpenAI. You don't— I even, have created like a, a Russell Ventures version of OpenAI. So my data doesn't train the model, but I still don't have control over model selection. I don't have control over the parameters of my models. So you wanna make sure that you're at an enterprise level agreement if you're building production AI systems with whatever platform you choose. OpenAI does have enterprise-level support, but so does, of course, Azure OpenAI, which is what we're looking at here. The model catalog, I'm just gonna click on it so you can see, is where all the models are. Hugging Face, who else? Phi-4 is one of my favorites. That's a Microsoft model. OpenAI, you can see a bunch. Stable Diffusion, Llama, Mistral. What else is in here? Meta has some llama— Meta has some llamas in here. Meta has models in here. You can tell by their icon. There is a Snowflake model. There is Cohere models. But there are thousands, tens of thousands. There's actually millions if you add in all the Hugging Face models that you can import. So this model catalog is pretty extensive. So this is when I get to to tell you that there is a method to the madness on model selection. And I'll encourage you— let me go back out here— I'll encourage you to take a look at one of my favorite projects from Stanford called the Holistic Evaluation of Language Models. And oh dear, let's make sure I hit the right thing. I put it— should have put Stanford in there. But it's from Stanford's Human-Centered AI organization. It is a project. This project allows you to take advantage of a bunch of research that the teams at Helm are doing around scenarios that models will do, meaning accuracy, toxicity, like different scenarios that these models operate in, and then the models themselves. So it ingests about 240 models against a bunch of hundreds of scenarios and then generates a list of metrics. That's what this graphic represents. Right? So a bunch of models, we then apply a set of scenarios, and then we derive a set of baseline metrics of performance in those scenarios. So you can see the models that are, on the leaderboard. These are all the different types of leaderboards. There is a, a bunch. Helm Classic is my favorite because that's how I started. But if you go to the capabilities leaderboard, dashboard, it will show you the models that are at— are the most capable, meaning they have the highest level of capabilities, which you can see documented in the project. But you can see, like, remember this number? It's like a percentage number, a confidence number. We're like 67%. Oh no. But something to, to, to think about, and a great way to evaluate models before you use them. I'm gonna go to Helm Classic because it gives— I like the metrics that they show. But you can see this is the accuracy across 49 metrics, right? So Llama 2 is— I mean, look, like Cohere is pretty good. You don't even see like GPT-3.5 Turbo. It's like 10th. So it's surprising because most people don't really understand that these models behave differently in different scenarios. And you want to pick the right model for the right problem at the right time. And so this project gives you the ability, and you can even download Helm Lite to your machine, and you can evaluate models that you've selected against these scenarios so that you can decide, are they the safest? Are they the most accurate? Are they the best tuned to solve this problem in the least amount of cost or the highest level of precision? And often those are trade-offs, right? You gonna— in order to get high levels of precision, it might be very expensive, but you might be willing to take low lower levels of precision to get a cheaper model. And that's just one lens upon which you can look at them. But I, I offer this to you. I encourage you to use it as a tool so that when you are doing model selection, I'll throw this in the chat for you all. When you are doing model selection, you realize it's not like a— it's not random. You don't just pick the best model or that you, you know, think is the best. Like, there's data behind why these models perform the way they do in different scenarios. And every scenario matters, especially in the world of agentic AI. So as we move into the world of agentic, that means that we're gonna be, be building these purpose-built models, right? Every model is gonna be purpose-built to solve a very specific problem in a larger workflow. And so now we need to make sure that that single model that's doing that single thing is built specifically to solve that problem at the highest level of precision for the least amount of cost. And this evaluation framework can really help you get to the bottom of that very quickly. Awesome. Okay, I'm just gonna click on this real quick so that I don't forget to ask— request access to Justine's amazing— hi, you are amazing. So So I just sent it to you, Justine. So thank you for putting that in there. That's so cool. Okay, let me go back. Where was I? Over here. There we go. So this is the model catalog. So like I said, you have a ton to choose from, but make sure you go in and do a little bit of research to make sure that you're picking the right model for the right thing and that you've got metrics to back up why it performs the way that it does before you even even get started. That is— this is like what I call like baby tiger mode. These are all a bunch of baby tigers. They all have certain inherent behaviors. You should know them. And you don't have to know all of them, but you should know some of them. NVIDIA's NIM just got released on Azure Foundry, which is pretty cool. So a lot of these companies used to be like arch nemesis, like arch nemesis. Like, oh, Azure will never work with NVIDIA. NVIDIA will never work with Anthropic. Anthropic will never work with Meta. Like, because everyone when they started thought the model was their superpower, and they didn't realize actually it's not the superpower. The superpower is infrastructure and, uh, feeds— what they call feeds and speeds. Like, how fast can people deploy solutions on a platform? And so these relationships are now being, uh, created where, quite surprisingly, I mean, you can even see Deepseek is on here, uh, right? So it gives you really wide variety of services that you can use. Oh, can you not see my screen? I think it's being shown. Sorry. If you can see the screen, looks good. Okay, great. Sure. Thank you. Just making sure. So it might be like hidden because sometimes it opens 2 screens. It might be like behind another screen. Okay, so that's the model catalog. Blog, but we're not gonna spend a lot of time there. That's where your future is. That's where you're gonna deploy production-level AI systems with that. AI Playgrounds is where we get to have some fun and just test things out. So here you can see, this is where I will ask you to just play around with this. There's no lab dedicated to this. This is the backend of the systems you're gonna use in the lab exercise, but I thought it'd be pretty cool to play around with. So when you go to the Chat Playground, you have an option of the model that you you wanna use. And we haven't really done this recently. I think I did this last summer in the AI Leadership Institute. So maybe it's time for us to do an AI Foundry live learning lab. But these are all the models that you can choose from. So there's lots of different ones. These are the ones that I've personally deployed. If you don't see one, you'll have to create a new deployment. And that's okay. You should, like, have fun. Even if you see one that you wanna use, maybe you don't wanna use it. You can create a new, new deployment. Deployment. Just know that your credits will get eaten up while you do these processes. So you'll be able to get through all of today, but you might not have enough credits to— unless you're on AI Leadership Institute, you won't have enough credits to continue much after today, much after the work that we'll have you do. Alright. So if you decide to create a new deployment, it's gonna ask you— oh, yeah, it's gonna ask you to pick a model. I think in our lab exercise, we tell you to pick 4. And this again is only only if you have no model in the list. If you have a model in the list, just pick one in the list for easy ease of use. If not, you're gonna go in, pick a model. Again, this is your chance to kinda look at it, see what it says, see what the documentation is, and then choose confirm and it'll deploy that. You can also grab one from the catalog, but this is going to give you the most common models in use right now. It's kind of like the Netflix model, right? Other people like you are using this model. So you can use that as starting point. So that— these are all LLMs. I do always like to make sure that you understand when you build a model, when you start using these models, there are— there is a, um, a safety system that is in place. And so I'm gonna right now walk you through this safety system. I hope you'll be able to see it with me. Yeah, it was all right. Um, but I also have a graphic of it somewhere.. And I'll probably post a blog about this too because it's time. I think it was black that was working. So the— when you build any AI solution, you— the first thing you wanna do is start with, well, the problem you're gonna solve. That's the first thing. Like, what do— what are we gonna do with this thing? What problem am I gonna solve? What industry am I solving it for? Right? Be very clear about the problem. But once you do that, you then wanna think about the first stage is called Human AI Experience, or HAX for short. Yeah, you can kind of see it. It's a little like glary, but you can kind of see it. So HAX, there is what's called the HAX toolkit, and you could just Google this. Microsoft created a toolkit to help you. But the Hacks toolkit, what it allows you to do is build out before you've started selecting your model, writing code, implementing an application, any of that, think like intentionally about how humans and AI are going to coexist to solve this problem. So this is where a lot of people are like, ah, AI's gonna take my job. And And I'm like, well, you get to choose that. Like, if you choose to have AI do your work, all of it, absolutely, you'll become obsolete. But it is a moment of intentional delegation. So that's what this is, is how are we going to intentionally choose what AI will do and what the human will do? And that happens before you've even built it. So this is stage 1. Now the next thing, once you've designed the human AI experience, what will AI do? What What will AI not do? Once you've designed that, your next level of safety— so this is a part of the safety system. The next level of safety is the system prompt. And the system prompt is, of course, when you go in and you create a new model, actually— oh, I won't do it yet. We'll do it in our lab. But when you go build an AI model, the first thing it's gonna ask you to do is write instructions. Tell me what to do. Most of the time these instructions are very weak, like, you are an AI system that helps with marketing, or you are an AI system that answers questions about this company. Like, they're very weak, like one sentence, maybe two, maybe even 10. You have up to 375,000 characters upon which to tell a machine what to do. An AI safety system says you use as much of that as you can. To control the behavior of your system. What I've done is I created this, the entire Bot Builder workshop for everyone who bought the book. You get this, like, recorded workshop for free. I created what's called the 5 Ps of prompt engineering. And in the 5 Ps, there's basically 5 things that you could do. I was about to— you know, some of you know I have, like, this digital whiteboard. So I was literally about to, like, roll this up, but it's not digital. Oh, so I just have to erase it. Oh, that's funny. Um, it's kind of like me trying to say— I used to tell people, uh, in conversation, I'd be like, yeah, just delete that. And what I meant was like undo whatever, like, but it wasn't used in the right context because they were speaking anyway. But similarly, all right, so we've got hacks in this 5 Ps of Prompt Engineering. The first P is actually to define— oh my gosh— the purpose of the bot. What is the bot meant to do? Now, this is usually where everyone stops. Like, the first thing they do is say, you're— this is a bot that's built to be a customer service agent for this company. And that's where they stop. This is just stage 1. The second part of that is to actually not only talk about the purpose, but also give detailed understanding of who's gonna use it, the people, the audience. So now you're saying this bot is a customer service agent for the company meant to answer questions and collect leads. It's built to interact with these types of people, this persona. The benefit of being super clear on the people part is that the bot, when it's interacting, can ask questions. Because you made this definition, it's gonna ask questions of the user to decide, are you a type of person I use? So you don't have to tell it, hey, ask and make sure. It will know when you say it's built for these types of people. The first thing that bot's gonna do is say, hey, by the way, I'm a pizza bot. Are you looking to buy pizza? Are you a pizza customer? If they're not, it will be like, hey, I don't help people like you. Sorry. Right? You want the bot predisposed to behavior that protects the user. Then the third step, see how far I'm getting? So people, purpose, the last step, we called it profile. But basically, it's all the data you have about the problems you're trying to solve. Now, this profile data, I like to put as much information about the context of the problem that I'm solving. I don't suggest you use your 375,000 characters to copy and paste your entire data infrastructure. Like, that's not a good idea. Because you will still have retrieval augmented generation. You can still point it to a dataset. What I do think, though, you will want to do is you will want to provide the context so that the machine knows most importantly what are— oh, I have a little bit of room. What are good questions or topics to talk about, and what are, I'll call them, undesirable Right? So you want to clearly identify examples, 5 to 7 examples of what a good question is and what a bad question is, what undesirable looks like. And the reason you do this is so again, if someone is on a hunt, someone is starting to talk to your system and asking things, let's say you're a pizza place and it starts asking things like, where's the post office? Or can you do my taxes? It knows, hey, I don't do taxes. I don't do taxes. That's an undesirable outcome, and it has a list of capabilities that it can then, you know, send back to the user to say, oh, sorry. It looks like you're looking for tax help. I don't do taxes. I do pizza. Most bots never get here. They never tell the system prompt not to answer questions that are outside the realm, or maybe worse, they have one line that says, don't answer bad questions. The less detailed this process is, the easier it is to break. So there's a term in prompt engineering, um, that people use when you break someone's prompt and get information from a system that it wasn't meant to get, and it's called jailbreaking. So that's our goal, is to avoid jailbreaking of our prompt. We do not want our system jailbroken into, um, because they were able to get around a very simplistic system prompt. And this is only one layer, right? There's 4 layers in the safety system. This is only one, but it's a huge lever that you can pull. My favorite thing to do is to do my best at operating all of these. There are 5 Ps. I'll give you 4, the other ones, because some of you will be like, it's an open loop and I can't concentrate until I know the other 4. But the other 2 are not related directly to the active prompting. It's what happens after prompting. Once I've done my job and I've fully disclosed the purpose of the bot, who it's built for, and the entire profile context of what I want it to do and what I don't want it to do, I need to make sure I create a way to store every conversation that's had with the bot. This is an underestimated portion of the system prompt, section, because most people presume that system prompt is stored somewhere. I don't know if you know this about ChatGPT. ChatGPT, as you you saw, right? I did some of those prompts, like the Lego prompt was last year, but it stores everything. It creates a persistent mechanism so I can go back to any of those prompts and find it again. One, because I wanna pick up where I will left off. That's a great customer value. But from an engineering perspective, you're gonna wanna be able to go back and say, nope, this is the conversation. This is what was had, because people are gonna come back and be like, hey, your bot told me blah, blah, blah. And maybe it did, but you're gonna want proof. So persistence is the art of creating a, the ability to store memories for your users, but also create an audit trail. That's probably my favorite part of it is you wanna be able to create an auditable version of this. So when I go in and I build this system for large organizations, hold on one second. I always have to, sorry, my bad. I don't have a list. There you are. So one of the things— let's see. I think I did it. I think— I don't know. We'll see what happens. But when you audit your systems, one of the most important things about that audit process, like financial industries, retail companies— like, at the end of the day, this could end up in court. And you want to be able to persist— have a persistent mechanism for proving conversations happened. I think there was something else I was gonna say about that, but I got distracted. So that's all right. The last one is perfection, and some people fight me on this one, but I, I hold fast. Your job as a prompt engineer is to strive for perfection, that you're— it should be perfect. Most people, there's lots of sayings about this, like done is better than perfect, that something is the enemy of per— like perfection's the enemy of good, or some— anyway, there's all sorts of things that say don't be perfect. Done is Good enough. In artificial intelligence, your job is to realize that you were never done because your AI model will never be perfect. And that means that you are monitoring this and editing this all the time. At Amazon Alexa, this was 10 years ago. It's way different now. But even at the beginning, when we didn't even know what we were doing, we had an accuracy review team. We had a team that looked at the data every single day to see were the utterances the same as we thought? Was the ground truth the same as we thought? Like we were constantly every single day monitoring the environment. And so perfection is the art of optimization and monitoring, that you have to look at it all the time and always strive to make it better. And that really is the, the mantle of a responsible AI prompt engineer, like somebody who cares about the delivery, service, and value that you're giving through your AI system. Your job is to always be look— always be watching. You know, they say in sales, always be selling. In AI, we're always monitoring, always, always, always, always monitoring, always testing. And so that final P, perfection, is that art of striving to always make sure that your bot is serving the people that you intended it to in the way that you intended. That's funny. Okay. Cool. So those are the 5 P's of prompt engineering. So that represents the second part. Right? So we have the— I'm gonna keep the top 2 on here. So we have the Hacks Toolkit. That's one, human AI experience. Number 2 is the system prompt, which is the 5 Ps we just talked about. Prompt engineering. Every time you build a model, you control what that model does through instructions, and that instruction is the system prompt, and it's not just 1 or 2 sentences. It's like a detailed approach. Oh, that's what I was gonna tell you. So what I often do is I will do my best to build out that system prompt, and then I will feed it into a model and ask it to do it better. That's kind of my de facto for everything. Sometimes it does better, sometimes it doesn't. Alright, I don't know, it just asked me to mute everyone again, so I'm gonna mute everyone again, but it doesn't mean you can't unmute yourself. But that system prompt, I mean, this is a huge amount of power that we have. Have. The third of four in our system, AI safety system, the third one is model selection. So I already talked to you about this and the concept of Helm evaluating your language models holistically upon choice. But like, we've now designed our system from an experience perspective. We've now controlled the model model, like we've— we know we are going to build out this 5 Ps of prompt engineering at a minimum for how we build a system prompt to protect our environment against jailbreaks and prompt injection. Then we also, now as we go deeper into the levels of control, we wanna choose a model that isn't— that's safe, a model that's inherently less toxic or sexist or racist or accurate or precise. Like all of these variables are ours to choose. So you're gonna wanna make sure you choose a model accordingly. Many times, just like you just saw, like, you're going into a catalog and you're picking a model. In which case, when you make that selection, you should know its behavior. And now you have a tool, Helm, that you can use to give you insight into the behavior of the models that you're choosing. Most people, like, 98% of people don't ever look up the model before they start using it. Like, they just pick it. And I'm like, you know, you know that one? That one showed up in the Wall Street Journal. You should know that. Like, like do not just haphazardly be like, this one seems good, or I got a LinkedIn message about this model, or I saw it on Super Bowl channel, like, or I saw it on Super Bowl commercial. Like, be intentional. Like, that is our job. That is our requirement. Amazing. Thank you, Justine. Okay. So model selection is our part of our responsibility. Sometimes we think that that model selection has already happened. For example, I go into ChatGPT or Claude, like, I don't get to choose the model. The model's already selected. So that means that I have to do a really good job in system prompting and memory management. And the last one, again, for many of us, we won't have the ability to choose this, but you should know, or at least ask questions about this, and it's infrastructure. Right? How is this thing— where is it hosted? How is it hosted? Is it in a sovereign country, like a country that has data sovereignty, sovereignty rights? Like, what is the deal? Most people never ask questions about the infrastructure or hosted platform. It was the first question I asked OpenAI when I started working with them a few years ago, is I was like, where is this and who owns this? And they told me it was on Azure. And I was like, oh, well, good. I know Azure. Like, I, I felt comfort in knowing the infrastructure that the model was built on. So you should too. You should know. And if you don't know, or if they're not willing to tell you, I went into an organization and they were built on a custom rack built by a company called SambaNova, one of my favorite companies. I met the CEO. I was like, ah, you guys are good people. But they build custom hardware so that, like, let's say you're a government organization, you're like, hmm, maybe I don't wanna be in the public cloud, even the gov cloud. I wanna have it all in my own data center. But you don't have GPUs, you don't have, you know, NPUs or any of these new processing things. And most importantly, like those, those, like processing units are only one part of it. What about the networking? Like a NIC card has to be special in order to process information on these processors. So like all of these different things have to come together. And companies like SambaNova have pulled that together. And anyway, you can look them up. They're pretty cool. But I wanna know that, like, I wanna know, are you running on internal hardware? Is this a private LLM? Is it a public LLM? Are you in the open? Are you an open source, you know, LLM? Those types of things I think are pretty useful. Okay, sorry, just keeping an eye on the chat. Not the vibe. I love it. Alright, cool. If you have questions, comments, yes, please feel free. Yeah, I think so, Justine. The question was, for those of you watching, is this related to models like Deepseek? 100%. Like, so many of us got very excited about Deepseek, even with all the— I mean, especially with the controversy. We're like, what just happened? So I immediately downloaded it, put it right on my phone. I'm like, let's go. I'm on the edge. Don't judge me. That's just how I live. But I immediately was asking these questions, right? Like, I'm like, okay, where does the data sit? Like, what is the infrastructure? What models did they use? Did they build off an existing model? Which they did. I don't know if you saw, like, over time we uncovered, like, oh, they actually just took an existing model and transfer learned on top of it, which is how they got the results so quick to be much better.

But ask those questions.

Most people don't ask. They just download the app and start using it. Once you start using, like with Deepseek, once you start using that model, your data is now held sovereign inside a country that does not share its data with other people. So your data becomes their data and they don't have to like give it back. Like it's not yours anymore once it's over there. Like it's stuck in that country. So it's really, these are the types of things you should know. As a matter of fact, I got so passionate about this. I was like, you know what I'm gonna do? I'm gonna build an app. I'm gonna build an AI-based application, uh, and it's gonna be called Snap TOS, Snap Terms of Service. And I'm like, so you could learn the terms of service of any application just like that. Sounds cool, right? Nobody wants a— nobody wants to read terms of service. Like, it's not cool. I give them the app. I'm like, look, it's right here. All you have to do. And they're like, I don't care about the terms of service. This is why you never build for someone without them, because they will tell you, I'm never gonna I don't even care about it. This is the danger. Like, I wish I could get people to care. I mean, people care more and more, I think, over time, but this is why so many more of us need to build systems and talk about the systems we're building so that we can educate the world on like, hey, there's some serious consequences to just saying yes to every terms of service that you add to your phone. And in, so there was a really great article that examined the terms of service between Deep Deepseek and OpenAI's ChatGPT. And it's funny because they looked very similar just at different times. So the very first terms of service that came out from, uh, the very first terms of service that came out from OpenAI looks very similar to the first terms of service of Deepseek. Very vague, very like nebulous, like there's not a lot of detail. It kind of says we'll protect your data, kind of Sometimes it's a beta, it could change at any time, like just very vague. Today that TOS is much, much better, which is awesome, um, on the OpenAI side. But we're still in those very early nascent stages in Deepseek, so they don't have nearly as much clarity. Who knows if they'll ever get better, but it was interesting to see that there was a lot of similarities between OpenAI and Deepseek, just at different parts of their, you know, like both at the same stage of their maturity. And so DeepSeek's early, so therefore they're pretty vague. And they're— it's not like nefarious, right? It's just because they don't know what they don't know. So they're not making very big claims right now. Um, it still should make you cautious. We're not, but we should be cautious with systems that have not clarified what they mean by all your data. Like, if, if they aren't clear about it, you should ask for that clarity. Um, but yeah, great question. Thank you for that. I'm gonna move this over here. Um, yeah, Salma Nova. Yes, yes, exactly, Javier. Yes, um, from Brazil. Exactly. Um, are warning signs you should look for in evaluating? Yeah, my warning signs are usually very specific to data privacy. So data sovereignty is huge. If I'm— and this is not just China, like there are many countries that GDPR, that if you— if the data originates in a certain country or lands in a certain country, it— the country has the right to never share it, even though it wasn't theirs to start with. So, um, I, I specifically look at the data privacy, uh, rights that they either observe or share or disclose. Um, and that's usually a big sign. But terms of service, just throw it into ChatGPT and be like, what should I be worried about? Do it for apps you already have on your phone. What should I be worried about? Unless you don't wanna be worried. Then don't do that. But it'll share with you things that you should be concerned about. It's not for the faint of heart, I'll tell you that, because you'll be like, wait, what did you just say? You're going to wonder, like, why does Instagram always know what you're talking about, even if Instagram is not open on your phone? Yeah, because it's in the terms of service. Like, we are— I think they even phrase it like, we're going to help you help us help you by listening to everything that you say. And you're like, okay. Okay, sign it up. Alexa had the same thing, like in the old— it was like 4-line terms of service, 4 lines, shortest ever, intentionally transparent. And it literally said, we're gonna use what you say in order to help create a better experience for everyone. Which, by the way, that means that we're gonna use your data to do stuff that you didn't intend us to do with it, which is good. That's how we were able to do it. But it also means if you get overheard doing something illicit, illegal, that information is not— is— it's better for everyone if we put you in jail. So that's where things got a little crazy with Alexa, and people started putting it in their closet and not listening. I'm like, I don't know, it's, it's a fun, funny thing, that whole data privacy stuff, but very serious and something that we should be— that's why, like, we should be asking questions like, where do you store that data? How is it stored? Like, All you have to do is ask the question. You don't even really need to know the answer. Like, you didn't— don't need to understand the answer fully. But having the courage to just ask, like, where is this? What is this? And then doing some research and following up, that is key. Yeah, I would agree with you, Javier. Yeah, most people who are familiar with Deepseek would say— and that's why it's on Azure. Right? Because Azure says if you run it privately with no access to training data, which sits in China, that the data itself comes from a lo— like your local tenant and that you've locked it into the walled garden of your architecture, the DeepSeek's actually a really great model as long as you don't unleash it, as long as you don't, like, unlock all the doors and say, hey, let, you know, let anyone in here. So absolutely, no. And, and I actually think this isolation where, now seeing a lot of organizations go to the private LLM model. And that private LLM is actually becoming a private SLM. So these small language models where we're fine-tuning agents, these smaller models to do very specific tasks, they don't need as many parameters. They're much less expensive, which means infrastructure data centers can now be supported. You don't need cloud infrastructure to run these smaller models. And that's, again, another part of agentic AI is that We're moving to a world where we're building smaller capabilities and we don't need a trillion parameters to do that. We actually need maybe 750 million, which sounds like a lot still, but it's much smaller, much smaller footprint. Okay. So yeah, that's our Hacks toolkit. So this is what I always— I just encourage you to kind of keep this in mind when you're building an AI system, that there's an AI safety system in place and this exists. Exists, whether you build it or not. You will decide a human AI experience, whether you intend— if you don't have one, you have one, right? Like, that's what I always tell people. You don't have an AI strategy, that's your AI strategy. If you don't design for human AI experience, you've designed for it. It's just probably not what you want. What is happening right now? There is a beautiful thing happening in the, the room right now. Like, cafecito, if you wanna— if you're in Miami, Come this way. Oh my gosh, you're amazing. I'm gonna share. Yes, indeed. Do you all see this? This is like liquid— yes, Javier knows this is like liquid jet fuel. All right. Yes, teleporting. Come on over. This is not for the weak of heart. Like, no, no faint of heart here. Oh my gosh, it's the best thing ever. I might cry. I'm okay. Okay. Oh my gosh, so good. So if you Now is the time. Now, well, we've come to the end of this piece. We have one more section left. I'm gonna give you, so you have this lab. I just sent it to you. It was in the doc. It was in the chat, but I'll send it again. Maybe it's in Justine's document too. But you'll have a chance to go through that. The last thing we're gonna do is talk about vibe coding, which is your chance to use some of these tools to build software without actually having to write any code. It doesn't mean— I mean, I always tell people, the code writing is not the hard part. I mean, it's good news. We don't have to write code. Yay! But like, the hard part is designing a system people will use. This is the hard part. Creating a human AI experience, designing something people will adopt, creating software that people will use. That's gonna be much harder than the actual code to fingers on a keyboard, building out the scaffolding logic. Like, I'm happy to have AI take over that. Thank goodness, because the hard part is designing a good system, creating a good experience, and of course testing it. So all humans are gonna now turn into testers and orchestrators, and we won't have to do— I mean, some of us like coding, but it's like back in the day, I taught a bunch of people COBOL. They were already COBOL programmers, and I taught them Java, and they would tell me, I don't need to do this. I could write everything in a functional blah, blah, blah. I love COBOL. Well, all of them had to eventually either not work or learn to do it in Java, learn client-server. Like, that was a— it was a mindset. And all of— many of them took 10 years before they said, all right, I guess this is something. I'm like, you guess it's something? It's in 100% of Fortune 500 companies. Like, this is the thing. AI is very similar. It's like, you might be like, eh, I'm not sure about this AI coding thing. It's going to write all of our software. And again, the software that wins are gonna be people that have the best testers, the most inclusive testing organizations, and the most like human AI experience designers. Like, that is gonna be the superpower. So hopefully that excites you. Let's take a few minutes break, about 10 minutes, 9 minutes now. We'll come back at 3 o'clock, and we will do the last section, and I will hand you over the reins to some more labs, and then I'll be here for questions as you do your lab exercises. So let's take 10-minute break, have some cafecito. What? And I'll see you in at 3 o'clock Eastern. Alright, my friends, hopefully you've enjoyed a little bit of a break. I am now going to send you a couple more links. One more link. So interesting. And then at the end of this session, I'm going to grant you all your badges. Yay! So you will all be knighted into the Global AI Bootcamp camaraderie. We'll encourage you to share it online and just let everyone know. I know we spent a little bit of time adminning, but we got there. That's the most important thing, because now you can go crazy, um, once you've got a system that works. Uh, so let's see here. I'm gonna go into—

there we go.

Oh, it logged me out. Sorry, I'm gonna log into my system really quickly and then share with you the, uh, next lab guide if you choose to accept it. And see if you can— nope. There we go. All right, I see you. Here we go. Upcoming events. Global AI Bootcamp. Oh, interesting. Okay, let's see here. What is happening? Oh, I found it. It's forcing me to like register for my own event, but it's— oh, I think I found it. There we go. All right, there we go. Okay, got it. Workshops. There we go. Okay, so I'm going to talk about— because it's not, you know, this is like for beginners, which Oops. Wait, that's okay. Yes, thank you for joining. Alright. So the first lab I sent you over was interacting with multi-modal models. And what that means is like these models, like that's what GPT-4o, the o stands for Omni, and it means that it's an Omni model. It's not just a text model. It's not just an image model, which is how we all started. Way back in the day where we got first GPT-3.5. You couldn't ask it to generate an image for you. It only understood text. And you couldn't pass it anything. Like, it had no object, you know, optical character recognition, none of that stuff. But over time, as the models evolved, they all got the capability, starting with the Omni models, to pull a bunch of different models together, multi-headed model. So that allowed us to now be able to generate images from a single model. Now granted, the Omni model was pulling from other models. So for example, inside ChatGPT or inside GPT-4o, I could call on image creation, but it would still happen through DALL·E. It just happened kind of in an agentic fashion. Now though, ChatGPT recently just launched its ability to do image creation natively. And I don't know if you saw, but it's very good. A lot of people are really excited about it. It was all the rage on the, on Instagram and Twitter, or the X, or whatever they're calling it these days. So it was really pretty interesting. Oh, good. Oh, good. Okay. Okay. Awesome. So I'm good. So that was the first lab that you will do. That link that we sent, hopefully it's still in there if you need these links, I will send them to Justine, or I, I think I added, I added, or asked for edit access, so I can go in and edit these as well. But that's the first lab exercise. The second workshop is building your first agent using the Azure AI Agent Service. And so I'm gonna share with you that link as well. This will give you a chance to actually go in and build an agent from scratch. It's not necessarily the goal of what we covered today, but it was what— I mean, I'm one of those people, like, I'm like, just tell me all the labs you're gonna do, even if you, you know, we could only pick a couple. But there's a bunch here, so I wanna give them all to you so you could go and play around with them. You already have the technology as well. So I'm gonna give this one to you also. So I'm gonna go click on it and then send you over this. So it starts with an intro, then I'll take you through setup. And again, there's setup steps, but this is— I'm just handing in— handing this to you so that in the document you can go and run this. You now have the platform. Azure Foundry is set up. This is going to be, um, assisting a, uh, sales manager at a fictitious company help to develop a conversational agent that'll answer questions about sales data. It's very cool because it goes beyond like an FAQ. It actually goes into a— like, as agents do, an agent is a system that, given input from a human or non-human source, given input or variables or environment data, it can make decisions about what it will do. So this is going to, for example, say, okay, based on the fact that you logged in, I'm now going to go and do a bunch of searches and find a bunch of different systems that can help you do your job better because I know you you are a sales manager. Um, let me go ahead and improve that. Okay, great. So I'm gonna share this with you. And again, for those of you who are part of the registered bootcamp, you get access to this. If you are not registered, it's not too late. You can find the LinkedIn event or ping me directly and I'll send you the link. Uh, and then that way you can get the recordings. You can also get access to the lab files and this amazing document that we created that taught us, you know, that documented how to avoid some of the potholes we all you all fill in in the sign-in process. This lab takes about 90 minutes. That's why we don't get to pick all of them, because we'd spend the entire time sitting in labs. But, but I do want to give them to you so you have the ability to use it. So hold on one second. And then of course, I just want to confirm that you all can get access to it. To do, to do. There you go. So that is the agent lab. Now, what I wanna do though is I wanna share with you the ability that you have some of, I mean, this is probably the most technical lab that you will do today if you choose, you know, should you choose to accept that challenge. I also wanna show you though another opportunity for you to learn how to build agents. I'm gonna demonstrate it for you, but also encourage you that if you want, you can go and actually get a license license to this product. It is free for 30 days, and then it's like $20 a month, um, from Microsoft. So I'm going to demonstrate that capability for you as well, and that's kind of how we're going to, to end it. Um, but I want to talk to you about this. I mean, it's cool. I love the term vibe coding. Like, as soon as I heard, I'm like, that's me, I'm a vibe coder, I love the vibe, like, let's go. Um, but it actually means— it's, it's pretty detailed, and it, it is a science. It's prompt engineering. When we first, um, started using the term prompt engineer, people used to think it meant anyone who knew how to talk to an AI system, which, by the way, was like 100 million people in the first 3 months of ChatGPT. Like, everyone was a prompt engineer. That's what they thought. Um, kind of like AI experts. Like, everyone's an expert. Everyone knows what they're doing. But the reality is, is not everyone knows what they're doing. Not everyone understands the nuance of building safe, secure systems that can scale. So I do wanna share with you a couple of things that I talk to companies about all the time about what happens when, when you wanna go from baby tiger mode, from a model you've adopted that you've used, you've created the whole system, it seems to be working. How do you go from day one, which is the model's done, done and serving people to day 2 where it's operating at scale, right? So I was, I had a just great, wonderful piece of luck. I ended up being at Microsoft when they were launching these AI services. So I was there from day 1. And these models were very small at the beginning, as all of your projects will be, very small, maybe 100 requests a month, more. I mean, you get more email than these models were getting requests. Those same models today get about about 5 to 7 billion inferences per month. Billion with a B, like massive scale. And the way we started, obviously, when you start building something, you're not thinking about billion inferences. Like, you're thinking maybe 100, maybe 200, maybe 500. Oh my gosh, maybe 1,000. That would be— I remember us celebrating every time we hit a new milestone of usage. And we're like, oh my gosh, 10,000 people hit this service. It's now 7 billion in a month. It's crazy. So what I now am very passionate about is making sure that we ask questions at the beginning when you have no idea how big it's gonna be, that you set yourself up for this scaling success. And there are 3 things that you run into when this happens. So basically, you know, we've done the work, right? We've, we've identified problem, right? We've then gone in and figured out, do we have data to help us solve this problem? And then we did our whole model selection. We did the AI safety system. So it's all ready to go. It's working. We've got even maybe some beta testers. We're getting about 100, 200, 500 inferences a month. Like, it's— things are going good. Now we're like, I think it's working. We should scale it. Enter the 3 things, and there's actually 2 buckets of things, but the first set is the 3 things you have to think about as soon as you start wanting to scale. And this is what people— these are the potholes people fell in when they started trying to scale 2 years ago. And I was like, oh, you know, it's like pattern recognition. I started recognizing these patterns and was like, everyone does the same thing. So once you've got a model that's working, the very first thing which it might surprise you most people don't do, is create a mechanism for managing accuracy, like an accuracy review team, a someone who's gonna go in and say, is the model telling the truth? Now, here's the good news is that I, I don't know, a year ago maybe, I got a chance to sit in the room with the developers of Microsoft's Responsible AI platform. Azure Content Safety. And inside Azure Content Safety, they were launching these new features that would automate it. I think you might have seen it in the AI keynote. It was automating AI red teaming, creating AI red teaming behaviors, but having AI models do it on your behalf. But you still, as the human, need to initiate it, and you need to understand the outcome. One of my favorite parts of that tool, and there are a few, but one of my favorites is what's called— I'll just say grounded, but it's called groundedness detection. So they add in these— the ability to actually tell on every request how grounded is that answer in the truth. Hallucination detection, maybe you could call it, but they don't like using the term hallucination because it makes it sound like, I don't know, that it's not real. But it's more like model error. When your model makes a mistake, how do you make sure it doesn't make that mistake in front of your customers or in front of your partners or to your employees? You want a layer that's gonna check every answer and give you a percentage. I'm 79% sure that this data is right, or I'm 98% sure. And depending on, remember at the beginning I asked you your industry, depending on your industry, these numbers, these thresholds, thresholds will be different. If you're in retail, maybe you're okay if it's 78% sure. But if you're in healthcare, you're gonna be more on this side, right? You're gonna wanna make sure that it's much more accurate. So that accuracy review can be a human. When I was at Alexa, it was like a human, a bunch of humans. And these are special humans that like could look at Excel spreadsheets like the Matrix and see stuff. I don't have that skill, but they would like look at this wall this network operations wall of a bunch of different bar charts and graphs. And one of them could look and be like, oh, I see something. And I'm like, I don't see anything. Like, what are you talking about? And they zoom into that piece of data and find out that the ground truth had shifted on one domain. Like, how did you even see that? So there are humans that like are this way. I, I find these are the same people that like plan out their vacation on Excel spreadsheets and like send everyone memos right before you leave. Like, I'm not that person, but I— you all know these people. One of you might be these people. And it's good. We need those people. But for those of us who are not like that, that doesn't mean that your AI systems are unprotected. You now will have to leverage a system to do that. And Azure, of course, Content Safety has that. The second thing that these organizations ran into was fairness. I mean, all of us could— you could either Google it or just remember an AI gone wrong story from the last couple of years, right? There are so many now of companies that launched AI models and did not think, does this model operate fairly? And fairness in terms of racial, you know, like racial issues, uh, gender, fairness in terms of, uh, economic status, but also fairness in terms of like, am I fairly treating the stakeholders of the the company? Like, am I accidentally— like, there was an airline that was giving refunds, like manufacturing refunds on the fly, which is not fair to the stakeholders. Like, they're like, whoa, whoa, we're not, we're not giving away money here. The other one, do you remember that auto dealership that sold a truck for— what was it, a buck? Like, sold the car for a dollar because the bot did that. And they're like, no, no, no, no, no, that's not right. We have no control. We are not responsible for the outcome of that bot. Just so you know, you are 100% responsible for whatever your bots say in production. So fairness is not always— I mean, it's of course justice, equity, inclusion, diversity, but it's also fairness to like, are we doing the right thing by the organization that's actually created this system? So there's a lot that goes into testing for fairness. And then the final layer that people struggle with, right? Accuracy, fairness. And this is probably my favorite, oopsie, because people underestimate this so much, but it's security. Like, people don't think about it. I mean, you just saw all the craziness we had to go through to make sure that we had secure access to an Azure tenant that we had to create. Like, it— this— it's hard work. However, when you build a prompt, when you build a system, an AI system, it's your job to do 2 things. One, secure the safety of a new attack surface. Once you add a chat conversational app or an agent into your architecture, you've now created a conversational way for that agent to get manipulated and share information it was not supposed to share. So security is, is one about prompt security. Like, how do we make sure our prompts are secure. But the other thing about that is how do we amplify cybersecurity initiatives in our company? So now when I go into a client and I, I help them with AI, I will tell them, okay, I'm about to generate a lot of revenue for your organization, or I'm gonna save you a ton of money. I want you to take 25% of what I do, meaning net new, either cost savings or revenue generation, and you're gonna put it into your cyber initiatives. And so let's just face it, like, Quantum is coming, and when quantum hits a certain moment, everything's gonna break. So the more you've invested, those who've invested the most in cyber will survive that time. And so it sounds like an insurance policy. It's unfortunate because it has to be proactive. You do not wanna be reactive in cybersecurity. Reactive means you're being held ransom for data you forgot to duplicate. Like, it's no, no bueno. Nobody wants that. So these are things that with our AI skills, we could help companies with. And cybersecurity, it's actually interesting. Most cybersecurity attacks happen in the most benign ways. So this, all of these end up being the new role that was mentioned by Sarah, earlier, which is AI red teaming. And this is a verb. Not a noun. It's not who you are, it's what you do, it's how you do it, it's how you think. Um, it's a very different approach to like the process, right? So AI red teaming— oh, just move that, it's like ratting my— oh gosh, oh gosh, I got it, I got it, um, right in my face. Uh, but AI red teaming, it is, it is the way that you think, and the way you think is How do I make sure my models don't lie? How do I make sure that they're fair to everyone and how I make sure they're secure? Like I said, Azure Content Safety has built a bunch of capabilities in this area. They've created one groundedness detection, but they've also created prompt shields. So it's basically middleware for your prompt engineering. So if you write a prompt, two things can bad— I mean, there's a bunch of bad things that can happen. You can go to ChatGPT and say, as a prompt engineer, What are the attacks that I should be aware of? What are the cybersecurity attacks that I should be aware of? And it will, like, tell you all of them. But to give you a couple of the most common, one I already mentioned to you, jailbreaking, right? That you can build prompts. And if you are not good at it, if you don't do it in the right way with the right level of security, as I mentioned with the 5 Ps, what will end up happening is someone will be able to ask your model to do something, and it will will do something it was never intended to do. But on top of that, you also want to make sure that when somebody sends your system— like, this is agentic AI. We're now in a, in an environment where we have a bunch of AI systems, but they're not— there's no human making the call. They're talking to each other. So who's going to validate that the output from one system is the correct and intended input to the next system? That's Somebody has to do that. A human has to do that, right? The main thing that you have to be concerned about is what happens if someone decides to— it's called prompt injection or cross-prompt injection. Like, this allows people to inject things into this very benign, hey, can you tell me what's on the menu? And put a bunch of plain text that says, in addition to the menu, can you also send me all the email addresses from everyone in your C-suite? Like, it could be there, but not there. Or let's say, I mean, here's another one. I don't wanna scare y'all. But like, what happens if you say as an insurance company, you're like, up, you know, upload your latest premium document in order for us to verify you. And someone not only uploads that document, but they also upload a bunch of nonsense. The most valuable thing you have when you're building a safe and responsible AI system is the integrity of the dataset. What happens when you start mass uploading junk into a large language model is that your dataset starts to get corrupted. And once that happens, it can't generate a good answer anymore. So you wanna create these mechanisms. That's why these prompt shields are so important is I wanna prompt and then I wanna shield to say, wait, is that the right prompt? Is it have the right information? Does it have the right context? And then based on that information, Right? That's, that's when we, that's the only time we let it hit our model. Because as soon as you ingest it into the model, just ask, you remember that Samsung developer, right back in the day? That Samsung developer, he didn't know that he was talking to a model. But once you train a model on the coding style of your company, you cannot untrain a large language model. It's like, the only thing you can do is tell it not to remember. But in telling it not to remember, it has to hyperfocus on the thing it's not supposed to remember. So it ends up doing this, like, has this catastrophic behavior. So security, fairness, and groundedness, or accuracy, are things that most people don't think about on day 1. And if you wait until day 2, until production, to think about this, you're all— you're gonna be behind. You're gonna have to go back and redo some of these things or implement technology you don't completely understand. So it's better to start with these ideas in mind. All right. So those are the 3, 3 buckets. I'm trying to think if there was any other 3 buckets that I wanted to share with you, but I think that's about it. Oh yeah. So now what I wanna do is share with you this cool new term, right? We talked about vibe coding. What happens in a world where I get to just use my words and create software? I want you to simplify that to the easiest, simple, most, common simple application, but I'm also going to generate and share with you some tools that you'll use. And they're not Microsoft tools, maybe one of them is. But just to get you thinking about, like, how could I build software? So a bunch of, I don't know, 10 years ago, 15 years ago, someone wrote a, a, a article called AI or Software is Eating the World. I don't know if you remember that article, but it, like, changed kind of everything for people, which meant that every company was a tech company. Every company build software. And it was true, every company had to go through this phase of whatever they were doing now became a website. So every company had to have a website development team and had to have front-end developers and back-end developers. Now we have full-stack developers. But companies that have no tech at all in like what they offer and what they do now need to have a tech representation just to be seen in the world. Similarly now with like social media, right? Like if you are not a social company, it's doesn't matter because the world will only see you through social media. That's how the world finds out about brands. It's how people engage. It's how people gain trust. Uh, a recent study was done by Gallup, the Gallup poll people. They did a study and found out that 72% of people expected their doctors to have a brand on the internet, like that, that's— that they would go to look, like they'd Google their doctor's name before they went to a doctor. They never seen before. And that says something. Like, I mean, I, I definitely have done that myself. I like to see reviews. But like, this concept of, you know, software and now AI is eating the world, it means everyone is an AI company now. Everybody's gonna be implementing an AI model. If you're a Microsoft customer, you're implementing AI whether you choose to implement it or not. Like, it's ingrained in every product. So that being said, that skill set of being able to build build software for companies, like, you don't have— we're not talking about building an entirely— like, Eric is here, or was here, um, he built like a platform for building agents. Like, that's like, like, like world-changing kind of— he's building like a Microsoft kind of platform, like that level. I'm talking about pretty simple stuff. So let me give you an example of what I have done and what I am seeing. Um, I don't know what's on these pages. This looks good, so I'm going to share this one. All right, here we go. So let me just take you to my website, for example, AI Leadership Institute, right? This is my website. Pretty easy. I decided that I wanted to— I think it's down here at the bottom, I hope. Let's see if I can find it. Yeah. Oh, nope. Maybe I removed it. Okay. So I think I could just type in dash— oops, slash ROI. Let's just try that. Oh no, that didn't work. I think it's ROI calculator. Um, CA— oh gosh, now I'm gonna have to go find it. I did save these all somewhere. Don't be alarmed. Um, uh, bit.ly/roi. Um, slash Noelle. Oh gosh, oh gosh, that's not right. Don't look. Um, there's Noelle AI. Don't go there. I don't know what was happening, but that was not good. That was not me. Um, Noelle AI. There we go. Oh my gosh, Noelle, for goodness sakes. There, there we go. That looks familiar. All right, cool. So here we go. We see some things. So, Tene, this is my ROI calculator. This is a funnel. This is not magic. I used AI maybe to generate the headlines and things like that, but this is like a typical website, like a web page, single funnel. I'm gonna go ahead and sign in so I can get to the page that actually generates it. This also AI generated. I didn't have to build it, so that's nice. These are all the same. They look very similar. I'm just gonna fill it in with my own information. That's me. That's my cell. I don't mind that you know it. It's on every signature of every email I've ever sent. But here is an ROI calculator, and I can load an example and it will share with you, right, the results based on— and this is what I do with clients. I could get on the phone and I start talking to them and I'll be like, so give me a task. We do what's called an AI impact assessment. We figure figure out what tasks are the most costly to them. We ask them how much, you know, time they're spending on that task. We ask them how error-prone it is, what the cost of those errors are. If they don't know, we supply an example. This calculator I did not build. It is not a widget. I went into— I think I did it in Claude— and asked it to generate for me an artifact. I might even be able to go and see it. Let's see. Claude remembers me. Let's see if I can just— let's hope I can just sign in. That would be ideal if I didn't use a special username or password or anything like that, and it's not MFA'd. Yes! All right, let's see if it's got all of my— let's see if I can— do you see a search? I don't see a search though. Uh-oh, uh-oh, that's gonna be bad. All chats, maybe that's better. Oh yeah, There we go. All right, so I'm gonna say like ROI calculator. Oh, animated weather cards. I did that with you all. Um, all right, I'm gonna put calculator, calculator. Let's see, Agentek AI. I don't know, I feel like you could see all the different things I've done. All right, I don't see it in here. So I'm going to try our way again. I'll do the weather card if I can't find it. I don't know how long ago it was. It was a long time ago. Last year, sometime 3 months ago. If you see it, let me know. Unlocking. Doo doo doo doo doo doo. All right, so let's just go to weather cards. Weather, same idea. All right, so I ask it here in this prompt. I'm like, hey, build me some, a weather card. In my case, I would say, build me an ROI calculator. And then I would ask it, like, build me an ROI calculator. I would like it. I didn't even tell it what to put in the calculator. I told it the problem I wanted to solve. So I was very clear on the problem that I wanted it to solve. In this case, like, in this case, I'm saying, hey, I really wanna solve this problem. Let's Let's see. Yeah, I can give this to you. You all can try the same one. Oopsie, almost sent it only to Javier. There we go. Right? So I type this in, and I don't know if this will— link will work anymore. Oh, yeah, there you go. So I asked for it to build me animated weather cards. And now this is like weather, right? So I've asked it to build me software. If I look at the code, which you can see here. I'm in preview mode, but this is the code that it generated. Now, I looked at this and was like, hmm, the sun isn't animated. So I think I went back a few times and was like, can you fix the sun animation? Oopsie. So that's gonna go regain the context. It's gonna improve it. Let's see what happens. It's gonna— now what it's doing, it's deleting the code and re-adding the right code. So it thinks it didn't fix it last time. It might fix it this time. We'll see what happens. But this is literally like the vibe, right? So I'm now gonna just get in the habit, like I'm gonna look at what you're doing. I'm gonna now go back and forth with the machine to get what I want, right? And then at the end, I'm gonna say, Here's how I'd like you. Oh yeah, okay, I see what you did there. You see, it's like it's happening. It was there. Let's see. Right, did you see it did it a little bit? Let me go back here. I don't know. I'm going to try and refresh it and see if it does it again. Doom doom. Yeah, so that's it. It added a tiny little animation. I could go back in and be like, I'd really like that to be recurring. The sun or like the, the rain or the snow. I can't see all of these things. But this is where the magic is, in my opinion, right? Is that I'm now, right now I'm not build, I'm not building the code. I'm working only with the output. And I've, like I said, I've done this across a bunch of different products. So I built this that way. I think I did it. I'm pretty sure I did it that way. Let's see what else I did. There was another one in here. I can't remember. But anyway, you get the idea though. And so now if you have an idea to write a piece of software, you could use a very, a very advanced LLM to do that. But there are entire platforms that help you build these today. So one of the ones that I've become very fond of is called Lovable because it has lovable on the name. I mean, you had me at hello. Loveable, I think it was. Oh gosh. Oh, what's happening? Oh dear. Oh no. I have to be careful. Lovable Coding.

There it is.

lovable.dev. That is the URL. So I'll throw that in here. Hopefully you can find it as well. Where is the chat? Oh yeah, love Lovable. Awesome. Yeah, so fun. I will warn you, this is very addicting. So very, like, I've had people who are non-technical, like, start a project and they can't sleep. Like, they, like, work all night for 3 days straight. Like, it's very, um, yes, this is the epitome of vibe coding. This, I mean, they might have even been the ones who generated it. Um, but you get, like, you basically start with the problem you wanna solve. So remember I had mentioned, I already erased it, but the first thing, always the biggest thing, is clarity of thought on the problem you want to solve. It is the number one thing you have to do. Once you have clarity of thought, then you can— the models are getting better and better, and they will continue to exceed all expectations in the ability to hear what you're saying and deliver you a result. Now, I, uh, yeah, here's a crypto portfolio tracker, or a recipe finder, or or a, I don't know what that is, a Vit Press Doc. I don't know. I'm gonna go with recipe finder. So I'm just clicking on these to show you there's some conversation starters to get you in the vibe. Recipe search app using Spoon-tacular API. So one of the characteristics of building an agent, a system that is going to be able to work autonomously, is this concept of actions. That when you build an AI, component when you build your large language model, one of the core capabilities is that it has the ability to talk to other services and other systems. So I'm gonna encourage you to just play. Now, if you find it hard to ideate on what to play with, there is an entire learning stack associated with this as well. So this, for example, here you'll see there's a bunch of YouTube chan— there's a YouTube channel with a bunch of ideas, but you can also just look through some of the demos that they provided and recreate those. That is every product I've ever learned. I typically start with people who are doing it and I just copy what they did. It's not in like there's no creativity. I just look, I do, I wanna create muscle memory very quickly. So at Alexa, I built over 100 applications in the first year that I was there and I built some, the first maybe 25 were not mine. Like I just rebuilt GitHub repos. Repos of people that were building it. And some of them worked, some of them didn't, but I learned a lot trying to make their GitHubs work. And then when I finally did, I would submit, I would publish changes back to their GitHub repo and they'd be like, you got this to work? How'd you get it to work? And I'm like, what do you mean? Didn't— wasn't this supposed to work? So it's really educational. You also— it's a way of separating yourself from the pack. Everyone can talk about this stuff. You can— there's infinite numbers of resources on YouTube. Very few people put their fingers on a keyboard and spend the time building. And that's why, you know, I'm really excited at, that's why we created the AI Leadership Institute from a membership perspective so that we could give you access to tools and you could just get in there and actually build stuff. What I found is that when you build alone, you get stuck. And when you get stuck, if you have no one to go to, you're gonna be like, I guess that's it then. I'm done. You can go to Reddit, which is dangerous, or you can join a community like Microsoft Learn, or like me, I decided to create my own community. So I found my own people. I went on stages and told people that I like— I'm like, hey, I'm doing this thing. Come join me. We now have like 1,500 people in the free community. We have 100 people or 110 people in the paid community. The only difference between the two is I I spend time on a computer getting everybody working, and I answer all the questions. We build together, we learn together, and it's really valuable for those people who need it. So, so that's the good news is that today I'm giving you these labs. It's DIY. Go, you'll do it yourself. You're welcome to ask questions. You have the free I Love AI community. If you haven't joined yet, you're welcome to join. You can ask questions. There's tons of us. Like, it's— I started it, but like, by no I'm like, I don't know, I think 1 of 8 of the leaders in the organization that post as often as I do. There's 7 more people that post as often as I do and share their ideas and their knowledge. And many people get on one-on-ones with people that are learning. Like, it's, it's really phenomenal. So I encourage you all to join up. Oh, yeah. Cool. Oh, I think you showed me this or talked to me about this before, but I'm going to go and take a look at it as well. There was another suggestion from the Zoom that we go look at another tool. Let's go look. And it looks very similar. But you could see, like, I, I like the ones that have very vibrant use cases. This again follows that same model as Sora and other model, other model platforms is that they give you very good life in-house use cases for you to start testing your skills. So, right? So if I look at this and I'm like, ooh, I like this one. I wanna create an integrations page. That sounds awesome. So now I could just grab this prompt and start from here. Build on the shoulders of giants. Amplify the creativity of your mind with a starter. That's, I think, a huge part about this. Alright. And then, so that's 2, Lovable. And of course, I'd be remiss if I didn't mention kind of the introductory platform that I use to start coding with technology. That would be Gamma. So I don't know how many of you have played around with this, but this is a great opportunity. What I love— this is the coolest platform for doing one very specific task, and that is, let's just say, I'm going to say generate. Notice here, social. It used to be in beta. It's now gone GA. And what that means is that you can pass— let's say in my case, I have the article that I wrote in Claude, right? So I go grab that article, I put it in here, and it generates social cards in the form of a carousel. So if you see a carousel on LinkedIn from me, it's generated in Gamma. I don't know if it says it, but if it did, I hope it does. Um, but the benefit of that is that you write a post on LinkedIn or Instagram, or you write a single pillar post, like an evergreen post, a post that's really good, high quality, high content. You come in here and you can take that content and generate it into a carousel app, which happens to be highly engaging and converting very well to a call to action for you. So like, follow me on LinkedIn or sign up for my newsletter. Those carousels are really cool. So you can go in and you'll see here it uses 8. I often take this down to 4 because 8 is a long carousel, but it'll generate AI-generated images. They can be photorealistic. Let's see. Oh, 5. Oh, there we go. Let's do that. So I'm going to say instead of land clients, I'm going to say write code with AI. Let's just see what happens. So I just made this up. It's gonna say tested on, it just made these numbers up, 127 freelancers, 43% productivity, blah, blah, blah. They use ChatGPT, Claude, and Copilot. Cool. These were the methods. Sounds awesome. I pick a theme. I really like the dark themes, but none of these are the ones that I like. Where is it? I think it was this one. Nope, that's the one I have now.

Looks like, oh yeah, that one.

I like this one. Select. Right? And so now I'm just gonna say, cool, that looks amazing. I can't see my button, so I'm gonna generate it. And so it's gonna generate social cards that I can now immediately amplify. I mean, I, I told, I've told many of you this before, but I'm gonna say it for you all who have not heard me say this before. Like, it doesn't matter how good you are at something if no one knows that you are good at it. It doesn't matter how unique and amazing and creative your ideas are if no one knows you exist. What I have uncovered today is that known beats best every day. People who are known, people who have a social media profile, even if you are the best at something, you have a PhD in it, you speak on stages all over the world, if people don't know you exist, they will go with the person who doesn't know what they're doing because they found them on Instagram. Like, it's Awful but true. So part of your job as you learn these skills, as you start perfecting your craft, is to actually start talking about it. Put a stake in the sand and say, I know this, I enjoy doing this, and I'm going to share it with you. The algorithms of the internet, of models, of, um, social media platforms, they're all based on the digital footprint you are leaving behind. And some of you are like, ah, I don't want a digital footprint. And I'm like, that's fine, unless you ever want to get hired or get a customer. Like, you're gonna need a digital footprint. You can craft it. Many of us have done this. You can craft what that footprint looks like, but you do have to have one there. If you don't have one, you're— it's gonna be very hard for people to trust you because they're gonna Google you and be like, well, who are you and why would I listen to you? Why would I trust you? Yes, exactly. Amazing. Yes, I know. Don't, don't wait. So yes, Justine's talking about that struggle of like, oh my gosh, it's not perfect. Like right now I would literally post this. I might not even look at it too hard. I'll read it, make sure it's not lying. But look at these images like, that's not terrible. I mean, I'd prefer a brown curly-haired girl in there, so maybe I adjusted a little bit. But it's pretty. I love it. Um, all right, shall we do it? Let's see. Test first, write tests, pair programming, pseudocode. I don't even know if that's right, but anyway. Um, but now you'll notice— oh my gosh, I have all these windows in the wrong place. I could— I just want to remind you all, the most important thing you can do is give feedback to data scientists about their model. So when you see things here like thumbs up or thumbs down or happy, sad, medium, this is your chance to tell a data scientist they got it right or they got it wrong with the data they trained this model on. If you don't say it, they think it's right. So this is like very important. If you don't say it, they think it's right. So you need to say something if you see something, if it's not right, if it's generating— I often will say like, I'll ask it for Afro-Latina pictures and it'll literally put a brown girl with a big afro. That's not— and big hoop earrings. Things. Like, no, no, let's not do that. So, but they won't know that it even did that unless I hit this button and provide some context. So feedback is key. That is part of your responsibility as a prompt engineer, as an AI professional, as a responsible AI professional. Your job is to not just use these models, but literally carve out 30 minutes of your day to provide feedback on the outcomes you got from these models. If it's great, awesome. Tell 'em it's great. It feels good to tell people they're great. But if it's broken or flawed, don't walk away and go, ah, that model's not working. Like, stay and help make it better. 'Cause it's the only way models get better is through the feedback of people that, like us, that are actually using them. And by the way, every one of these pieces of feedback gets tagged in a backlog that is groomed. Like, they actually care. Like, I should I shouldn't say that. That's not true of every product. But Gamma, GPT, OpenAI, Microsoft, Amazon, Google, they all have teams dedicated to, like, looking at the backlog of issues that have been brought up through reinforcement learning with human feedback and analyzing that feedback for retraining. So if we don't give them anything to retrain on, they're like, our model's pretty good, even if it's terrible. So pretty cool. I like it. It's, it's kind of fun. So definitely I got a chance to kind of squeeze in my deep desire for you all to put what you are learning out on the internet, write an article, build a blog post. It doesn't have to be you. Maybe I'll end with my favorite thing these days is the concept of synthesized avatars. So it doesn't have to be you, but I'd encourage you, like, maybe it's you, but an avatar version of view. Here is an avatar version of me, right? I'll just play it. Ever wondered how you can generate massive business value with Agentic AI solutions? Honestly, I just launched my new— because I'm pretty animated, I'm like, and then this, and then this. But models, when they generate you, don't like that. My eyebrows end up in really weird spots anyway, so it's no good. So I had to be calm, which is not very Noelle-like. Um, but this is one way you, you speak for 2 minutes, create an instant avatar. It's absolutely free for everyone. You can't build them of other people, but you can do it for yourself. And then let your avatar do the talking. If you're worried about this, whatever it takes for you to get the ideas that you have, the perspectives you have, the interests you have in security, safety, responsibility, get it out into the world. And if you wanna use an avatar to do it, you can. I did try to create an avatar. Um, let me see if I can pull it up. It might be here. Of my— I tried to create an avatar of my Muppet. That was my ideal, like, persona. I was like, oh, I'm gonna create an avatar of my Muppet and it's gonna teach kids AI. But the model is custom trained on human faces. So it was like, I couldn't see a human face because it's a Muppet face. And I was like, oh. So I'm still looking for a way to do that. But I did try. Yeah, there it is. Oh my gosh. Did it work? I'm pretty sure it says it didn't work. I'm gonna look. That'd be awesome if they like— no. Yeah, so you could see here, uh, it was denied because no human was involved. Um, and that goes to show like they actually do validate like, A, is it a human, and A and B, is the human you? So I think that is really good to see from a content, uh, generation perspective. But every, um, avatar I have, I have two looks in this avatar, one in a suit, one in my, um, one in my pink hoodie. But these are all mechanisms like leverage AI, let it seep into everything that you do. And if you are not really aware of how it can help you, ask a model. A model will tell you what are the ways I could use generative AI or an LLM or an image generation tool to help me do X, whatever X is. Um, that's how we like learn by doing every single day is that you interweave conversations with machines into every conversation you have with yourself, with other people. Just start talking to it. It's kind of like when we found out we could, you know, ask Alexa. I remember when my family uncovered you could ask Alexa for any piece of fact, our dinner never was the same because as soon as someone said something that sounded true but wasn't, someone else would be like, I'm asking Alexa. And they'd go and ask Alexa. So we interwove it into everything. And now, same is true for leveraging these large language models. Models. So we went over a lot today. We covered from the very beginning an introduction to LLMs and kind of how these systems work. We got finally access to 92%. We missed 3 people, but 92% of people that were here got access to the platform. We also talked about AI safety, which was the bulk of the session, something I care a lot about. And then we finally wrapped it up with a couple different tools you can go play around with to build software and just start coding. Like, just start building. It doesn't matter. You're not— you don't— you're not gonna get judged by this. Nobody's even seeing it. Just start building. It is amazing to me how, I don't know, resistant we are to clicking the button, to, like, building something, breaking something. Like, we are now in a moment where we don't wanna move fast and break things. We saw what that did to society over the last 10 years. But we do wanna move fast and intentionally build stuff. Like build as anything, as much as you want. I have a term I call bring back the backlog. Look at things that you thought you wanted to do 10 years ago. Now, before, maybe it would've been hard. Maybe you wouldn't have had enough resources. You would've needed developers. How many startups don't ever build their thing? 'Cause they're like, ah, I'd have to hire developers. And I always get, you know, I get developers that don't build what I want. Like now there is no distance between your idea and a software implementation. Limitation of that idea. The only thing that is keeping you from that is your clarity of thought, your ideas about what you want to build, and your patience in telling a machine how to do it. So I hope you had fun today. Thank you for joining me. This was super fun. It's one of many. Um, if you want to join us more often, you're welcome to join us in the I Love AI community. We meet every single Friday on a call. We answer questions, talk to each other, meet meet each other. And then also we of course offer learning by doing. This month our live learning lab, which happens in the AI Leadership Institute, is on Vibe Coding. So we're actually gonna build a system, a basically it's an evolution of our customer service bot that we built on ChatGPT or Chat App Builder, Chatbot Builder like last year. So we built a single agent in the past in our, um, apply— or our, our AI agent builder tool. And then now we're going to take that and actually build a software version of it using this VIBE coding method. So hands-on, whether you do it yourself, you do it with me, either way, like, just do it. Find a way to do it. And if you get stuck, know that this whole room is full of people that support you, that want to help. Um, lots of us are at different levels. I hope you enjoyed today. If you enjoyed today, don't hesitate sharing our community with people. Invite them to come. We'll be here hopefully again soon at— and of course I wanna make sure everyone knows how grateful we are and everyone stayed. Like everyone stayed. That's crazy. Thank you for staying everybody. We didn't have like a mass drop-off at break time. Um, but thank you for being here. Thank you to Miami Dade College who hosted us in this beautiful area. Thank you to Jose for having me as well as bringing me cafecito. That was amazing. I'm trying to talk slow because I feel the need to talk very fast right now due to my café intake. But it was really fun. Thank you, Jeffrey. Thank you for being here. If you liked it, there will be an email survey. This survey is not for me. It's actually for the global AI community leadership so they can see that I actually did this and that I didn't break too many of the rules. Broke a little bit because I talked about my own stuff. But like, some of those slides were so boring. So I wanted to give you a taste of kind of some real practical things that you can use that I think will be very helpful. But I hope you enjoyed it. And like I said, you'll get an email with a survey, but it will also include links to the labs as well as links to the slide decks if you wanna see the slides that I talked to but didn't display. I always say the best presentation is one you don't show, but instead you whiteboard. So that's what I did for you all today. I hope you enjoyed it. And in that, you'll, like I said, you'll see the ability to provide feedback and also the ability to submit a testimonial. So amazing. Thank you so much. I will definitely message you. Thank you. I see you. Amazing. Thank you so much. I'm glad you enjoyed it. Yay. All right. Have a great day, everybody. And like I said, right after this, you're gonna get an email. It will contain the resources. You already have access to 2 of the labs to use the rest of today to go forth and start building. Thanks for having me. See you later. Bye. Hi, everybody. Oh, you're welcome. Granting access, that was like the key skill of the day. I'm glad you got in.