I ❤️ AI Q&A: Real AI Trends, Shopping Agents, New Laws and Smarter Cheaper Models

Do you record the institute? I do. Yeah. Okay, good. Don't do it live today. It's recorded, but it's definitely live. Yeah, it was a good session. Hi everybody. Welcome. This is our Super Fun Friday. It's where we talk about, I'm actually gonna do a bit of a roundup. There's some crazy news going on right now, so I wanted to do a roundup of some of the most exciting stuff. Little bit of updates of what's happening in my world of I Love the, I Love AI community as well as the AI Leadership Institute. And then most importantly, give us a topic for discussing and then break out into some small groups and just discuss one of these pressing topics that I'll be discussing today. So welcome to this session. I guess it's our I Love AI community call. You're welcome to ask questions, raise your hand. If you don't know, everyone who is on this call, unless you're watching on YouTube or LinkedIn or Facebook, but if you're in the Zoom, that means you are part of our I Love AI community. We are almost 1,900 strong, so that's pretty exciting considering we just started, you know, last year. So it's pretty cool. A year and a half, I guess, ago. And this— that means that when you come on the Zoom, you can, you know, unmute, you can talk. I mean, everybody I've met and has joined this community I've become friends with. So you're welcome to join us on these calls. You can get your questions answered live. We will be answering 3 questions from our community members when you join the community, which you can do by going to noelle.ai. That's the website, noelle.ai. I really should put like a little banner or something on the screen when I say that. But, but yeah, noelle.ai. If you head on over there, you can join us either in the free community where you can join these calls and ask your questions, or you can join us in our premium community where you actually get access to tools and resources and probably most importantly, AI certifications, and I'm pretty excited about them. We just finished our latest one, AI Solutions Architect, and we have a couple of dozen people getting their certifications this week. So I'm really excited about all the work that has happened, and people seem to like it. So, so come on over, hang out, let us be your tribe. But that being said, I wanted to share with you a couple things that— oh, Hello, what's up? Oh, good to see you. I always like it when our, our, we have a couple AI bots that are saying hello in the chat. So, and I think maybe they're mine. Good to hear from you. Good to hear from you. Thank you everybody. So great. And I'll try and keep an eye on these on the chat so that if any questions come in, I can answer them. But everybody, when you join up, when you join the I Love AI community, everyone gets asked a question, which is basically like, hey, you know, what's, What's worrying you in the world of AI? What's the one thing you wish you could get some help with? And so what I'll be doing today is covering 3 of those questions. But first, let's start with a bit of a roundup over what kind of craziness is happening in the world of AI today. There's some really cool stuff happening. One, I don't know if you saw, but OpenAI and Broadcom actually created an agreement, uh, for custom AI chips. I think it's interesting because we're seeing a diversification in companies, big LLM companies diversifying where they're getting their hardware. And that's actually good, I think, for everyone. I don't really like it when everything's in one, you know, when we used to say things like you're an IBM shop or an Amazon shop or Microsoft shop, and usually that's just not always the best option to be all in one. Uh, it's great to diversify in your finances. It's great to diversify in your AI platforms. Uh, best of breed. So this, I think, is a great, great opportunity. Okay, cool. Another note-taker in the house. Some cool research that recently was released. One of them is a new model. Oh my gosh, you guys are going to love this. So there's been a couple of different levels of this research done over the last, I'll call it, 7 years, so maybe 8 years. So early in my time at Alexa, I was Speaking at, I think it was the University of Washington in Seattle. I was speaking at the University of Washington on Amazon Alexa and the power of voice. Uh, but there were also researchers there presenting their ideas on the impact of predictive text and what it does to a human's brain. And the research that they had shared was that the more someone used predictive text in, in email, right? Like Google, I think it was Meta actually that was, um, providing this feedback, but And the more you use predictive text, what they ended up doing in this study is they gave people predictive text for 6 months and then they took that predictive text away. Now, granted, this is in 2015, right? So then they took predictive text away and they analyzed what happened to that person's speech patterns or text patterns after that technology was removed. And what they uncovered, which has now been corroborated by more current research at MIT, I think, um, But they uncovered that we actually, you know, that over time the model trains the human, right? So the words that the model suggests— brilliant, fantastic, whatever— that those words, even if they were not in the vernacular of the human prior to using predictive text, that they could infuse new words into someone's vocabulary. And when that predictive text tool was removed, those new words would stay. That we as humans would think we came up with it all on our own, when in fact we'd been trained to use those words over time. Fascinating world we live in, right? So while we think we are using this AI, it's actually training us on how to talk, how to ask questions. So recently, um, FSU, Florida State— they had an amazing weekend, by the way— um, Florida State, uh, had, um, a— some researchers also found out that these large language models, they're not just shaping our text-based responses They are also shaping the way that we speak. So they analyzed over 22 million words from unscripted podcasts and found that words highly, uh, LLM favored words like delve, boast, meticulous, garner. What's another one? Um, y'all could probably think of words when you hear it, you're like, oh my God, when, when you see an LLM give that to you, um, what is one of it? Uh, there's one word that I, I actually have told my models not to use it, so I don't see it anymore, so I can't remember it. But, but there's certain words where you're like, Ugh, so LLM. Anyway, and so, um, this Cornell study, uh, from FSU uncovered that it's now seeing in the most recent podcast that humans are talking like they're LLMs. And it just reminded me of that study that came out in 2015. I was like, yeah, we, we know this. What are we going to do about it? Um, another thing to keep in mind, the, uh, 2025 AI Index Report. I remember this is the 3rd one, maybe. I remember when it came out last year. It is an incredibly cool resource. Actually, let me see if I can grab the link and put it in the chat for you all. Uh, here we go. Let me see. Uh, so this new index report, it came out last year. It feels like it was just yesterday that we got it. Um, but it shows some new details, some new enlightening details on the shift that's occurring across research, education, and policy. Um, and so Worth, worth taking a look. I will tell you that one of the secrets to my success is that I dive into the research that I trust. Human, uh, Human-Centered AI at Stanford is one of the organizations I trust the most. Um, and so I dive into the work that they have done. I research the papers that are used to create the study. Like, I just— this becomes the, the meat upon which I feed my community, my customers, my clients for the next year. And so this research study or this, this report that just came out is pretty good. So I'm gonna throw it in the chat if I can find it. Oh, there you are. No, let's see. Let me see if I can find the chat. There it is. I found it. I found it. And maybe someone's thrown it in already, right, Tara? Those— it's funny that you say that about the words Right, is that, is that people who speak— we, we actually in our previous call we were talking about the interesting part of neurodivergence and that some people, some people like myself, you know, if we're well educated, if we like came from academia, depending on our background, our lived experience, right, we are drawn to use things that now LLMs, you know, mass produce. So yeah, so it's interesting because They're now assessing that if you use those words, whether you inherently did or not, that you're LLMified. Oh, which is kind of sad. So, um, anyway, so there's the 2025 Index, something worth taking a look at. Like I said, there's a couple things. One of the things I think that I was most interested in, um, is the, uh, models, industry models that are coming out. This is exactly what I predicted would happen 2 years ago, uh, was that models became— as models started to get built by more and more companies, that the big investment would turn towards small language models, or large language models that were industry-focused, right? So medical models, manufacturing, supply chain, like, just very specific problem-specific focus. So industry is making significant investments in AI and are leading to pretty significant AI model development. One of the biggest, uh, changes. Um, it also does cite publications on AI, and, uh, US is still in the lead in publishing papers and research about AI. China in a second place. Germany— my husband's German, so, uh, Germany in third. So that's not terrible either. Um, but, but something to be thinking about. We're also seeing massive growth in the number of AI publications that are being generated, uh, in in the realm of computer science. So I like this report because it kind of gives you a sense of— and, and it's great to correlate this report, this, um, index report, to last year's index report and the index report the year before that. So you can kind of see— and all right, like, let's just be real, maybe we don't got time for that. So use NotebookLM, right? Upload all 3 reports and tell it to tell you a good story. Um, but use this information to kind of give you a sense of what reality looks like, right? Like, it's so easy to kind of get caught up in the media and in all the things that we're doing. So, so just a couple of things now in the legislation world. I wanted to give you a quick update on this. I saw this come out when I was in China, actually. But on September 4th, the First Lady, right, Melania Trump announced a brand new challenge. Now, I saw this, it was like in my feed, right? And I was like, presidential AI challenge, I'm all in. So I go to the website and I go to sign up and Well, I love the, I love the idea. It's not for me. It's kind, I don't know, I kind of want it to be for me, but it's for education. It's to really focus. So go look it up, Presidential AI Challenge. Maybe you'd be interested in, in taking a look at it. But it is, it is targeting education for 10 million Americans. However, not in, like, not us. Not you. Well, maybe some of you. Some of you might be in high school or in elementary school, but it's K through 12 AI education. So the challenge really is to encourage young people to come up with ideas and use AI to solve those problems, which I love. The other thing it's for is educators, which I also think is interesting because some of you know I'm a huge fan of the work that MIT has done in creating educational content that is free to use for teachers, for educators. And this content is amazing. Like, there's lab exercises and fun projects and proof of concepts, like, awesome. It's called RAISE, R-A-I-S-E. So you can go to raise.mit.edu and see more about it. But they ran into a challenge because educators— there was a huge gap between AI, cool AI content, and the educators who could teach it. Like, there was, There was no one in the middle who could help educators teach that content. So I tried to do it at a small scale in Florida, and I was pretty successful in Florida. But how do you scale that across not only the US but internationally? Of course, this presidential AI challenge is specifically for US educators and their students, but it's pretty exciting. So lots of people are involved. Google has donated. Microsoft has donated. I think Meta is involved. OpenAI is definitely involved. So something to look into and take a look at as well. Now, a couple other things. I always want to make sure that when I talk about legislation, I kind of get it accurately defined. But there are a couple of laws, or I'll say a couple of states worth looking into. I'm not going to talk about the specific legislation because by the time you hear this, it might even be different. But these these states are passing laws that are deviating from the AI— not deviating, but not— they're not waiting for legislation at the federal level to govern what they do at the state level. And there's lots of states that fall into this bucket, but I'm going to name a couple. Colorado recently passed a bill that delays the enforcement of its earlier AI safeguards law. So I don't know if you— should I say that again? Like, they passed a bill that's going to delay the enforcement of AI safety. And you might know why, right? This is not a bad thing, or I mean, who am I to pass judgment? But there is a compelling document called the AI Action Plan written by the federal government to complement the work that was done in the executive order by the last administration that says, hey, you know, let's, let's, let's be careful about being careful. Let's not use being careful as an excuse not to move forward, not to innovate. And while I appreciate the sentiment, it now means that the burden of building safe, responsible systems— that burden is not on legislation. It's not on a, you know, an administration of a government, right? It's on us as the people doing the building, doing the implementing, which makes my work so much more important, right? It makes all of our work being responsible AI leaders that much more interesting and important. So Colorado passed a bill that's delaying the enforcement of earlier safety laws in alignment with that, um, AI Act that is saying, hey, don't let, right, the— don't let forward progress be inhibited by safety protocols that you've decided to put in place. So they've delayed those safety protocols. It is what it is. So, so we are starting to see states, right, react to this and are in, um, this federal level, uh, in, in agreement. New York Um, there, a council member proposed licensing requirements for chatbot services, mandating disclaimers, accuracy alerts, mental health support links, session break prompts, um, to counteract psychological risk. Now, this I actually like. His name's Frank Morano. Um, like, he's proposing, like, hey, we are already seeing some dangers in using this— sorry, I have something in my eye— in using this technology, right? We're already seeing some concerns around mental health, around overreliance, around overanthropomorphization, right? Like where people are heavily integrating their lives with this technology. Maybe it's in our best interest as a council member in New York City, maybe we should add some restraints, add some governance, add some alerts, create some opportunity for the user to know that they're in kind of a, a failure mode, right? That something bad could happen based on the data we currently have. So that's kind of promising and interesting, um, to see. I always encourage you to keep up to date on these things yourself. You can definitely in research mode ask ChatGPT-5 to give you a sense of what's happening on a weekly basis. You of course can listen to these types of things, but I think it's cool. I mean, there's some, some really interesting things that are happening and they're only getting published in local media. Now I travel so much that I uncover these things like in New York or in Colorado, as I'm speaking at events. But it would— it's really— you, you of course have access to this as well. You just have to ask the right question, right? Where is this happening across our states? Let's see if there's anything— oh goodness, I don't even know if I wanna talk about the federal thing. So let's just go on to, to product stuff. So some of the things that I'm super excited in AI products is that OpenAI, of course, just did an XPU deal with Broadcom, which I don't know, some some of us saw coming, uh, because Broadcom had some really big changes recently. So that's kind of exciting. I wish I knew it early enough to make different investment choices. Um, but that's cool. Uh, I, like I said, as I started, I think that's a really great indication of diversification. I think that's great. Um, like I said, they're reducing their dependence on NVIDIA. It doesn't— I actually think that it's an, an addition, right? It's not like we're gonna move workload from NVIDIA to Broadcom. Like, It's just we're gonna continue to scale. How do we continue to amplify that scale? Now, in the study that Stanford released, they also talk about— oh, I actually took a picture of it. Maybe I'll throw it in the chat. They were talking about model usage, and it was a bit— I need to see where I would have put that. It was interesting. Microsoft commented on it earlier today as well. Let's see where I put that. Over here maybe. I feel like I downloaded it somewhere. Oh yeah, there it is. There it is. All right, cool. So I'm gonna throw this, but the AI model performance benchmarks are shifting. And so we're seeing everything happen faster, right? Like the training of these models are happening faster. The amount of, I mean, do you remember just 2 years ago we were like, we're gonna run out of data. It turns out in a world where we are generating more data every day, that, that isn't actually a major concern. I don't know. I, I feel like there's some researchers at Stanford currently looking at what happens in a world where our data that used to be solely created by humans. Now we're augmenting that with AI and human-created data. What does that do to future data ingestion? I'm sure there's some research. If you're familiar with it, throw it in the chat. Like, let us know if there's some research that I'm not aware of yet. I think that's pretty cool. Velocity, but the frontier of model, AI model performance is seeing, of course, this like hockey stick motion, meaning that performance is getting amazing while cost is decreasing. So I will, I'm gonna share this in the, in the chat for you all so you can see what I'm talking about. And this came from the study that I mentioned earlier. Oh, awesome. Thank you. Oh, I put that in there. Oh, good. Raise. Thank you, Sherilyn. Awesome. Yeah, exactly, Javier. Exactly. It does make sense to scale up. Yep. Why not? And you know, putting all your— this is why we have multi-cloud solutions now, right? We can't put all our eggs in a single basket. Just doesn't even make sense. Okay, here we go. I guess I'm uploading a file. Oh yeah, from my computer. Apparently I haven't done this recently. Let's see. I think it's a download. There we go. There it is. All right, so for those of you who are in Zoom, you can see that the image I'm referring to. And this is, like I said, it's actually good news. It means that things are going to get— that our things are going to continue to get cheaper for those of us who are using these systems while the performance gets better. And that is good for everybody, right? When models get cheaper to build and but the performance continues to exceed expectations, that's a really great place. For us to be. Let's see. All right. Was there anything else I wanted to talk to you about from a product perspective? I mean, it's always good to know that some of you might use Microsoft, some of you might not. But OpenAI, of course, as it launches its models, once those models are launched, the team at Azure AI, they test that model and harden it for enterprise use. And this is significant. I find that their time to market for new models, once that model becomes available on OpenAI, their time to market for productizing and hardening those models is crazy fast. It used to not be, it used to take 3 months back in like 3.5 and GPT-4, like as soon as 4 came out on OpenAI, it was like 2 months before we got it on Azure OpenAI. Now it's like same day, maybe the next day. So it's crazy fast. But one of the benefits of that is that it does this service, right? When you use Azure OpenAI service, which is one of the new models that they just launched, GPT Image 1, which is in preview, it actually can be— it is delivered as a service inside your cloud architecture. So all of the services, all the security, all of the role-based access control, all of the things that you have done to secure that tenant now will apply to the models that you are picking in OpenAI. And so that is one of the reasons why it's so compelling to use Azure OpenAI Service because it allows you to take advantage of all the work you did to build a good cloud, good secured cloud workload space, you now can just add your AI services into that collection. And for the most part, I think DevSecOps people are happy about that, right? I'd rather that happen, right? I'd rather take that model in as a workload that I manage in just like any other service in my cloud architecture than have you, you know, build a AI shadow AI implementation on some SaaS product that I don't know, right? As the person who owns the IT governance and IT, infrastructure. So the same is also true, of course, for Amazon and Bedrock, right? When you build models, you can just— you— they get to become part of your AWS infrastructure and get to take advantage of all the things you did to architect that solution, that architecture, that cloud architecture well. So I think that's everything. Oh, no, there was one story that came up. That I want to share with you. I got to be careful, I've been talking a lot today. But one story came up about shopping, AI shopping, AI in retail. Now I will tell you, I— you guys are gonna love this. I don't know, you might— at least a few of you who are my friends, you will love it. Um, some new people are gonna be like, Noelle's crazy. But I, uh, right now, because I lost my phone in— or I broke my phone in China I needed a replacement phone. Well, I don't even know if I brought them all in here. Oh yeah, there. So I have purchased lots of phones. I don't know if I even have them all with me. I have at least— oh yeah, I don't have the other one. It's out there. So this is the new Samsung something fold thing. I can't even, right? So I mean, I can't even help myself, right? I was like, oh, I need a new phone. Let me get a phone on a platform I have not— know nothing about. Like, I'm an Apple my, you know, I'm an iPhone user, uh, but it's pretty awesome. And, uh, it's got so many AI features. Uh, Samsung has its entire AI platform. How did none of y'all tell me about this? I do not. Anyway, so I spent the weekend like customizing every AI service on this phone, and I almost, dare I say it out loud, almost like it better. Um, I also though got the new Google Pixel 10, and it's It's gorgeous. I actually love the case for it because it's— the case is pretty, but the functionality also good. So what I do, and I do this on Amazon too, but like yesterday I was watching— my husband's into two types of shows, the kinds that make cars and like do a reveal at the end, kind of like houses but with cars, and then the cooking shows. And so he's watching Beat Bobby Flay. Any of you know this show? Anyway, Beat Bobby Flay, where they like compete Anyway, I like it because I'm into competition, but Bobby Flay is wearing a sweatshirt like this and it's got a huge glittery tiger on the shirt. And I'm like, oh my goodness, I have to have it. So I take my new Google Pixel phone. This isn't it. And I, I open up Google, which has AI mode and Google Lens, which has been out for a very long time. And I take a picture of his sweatshirt. And guess what happened? It showed me 45 different places I could buy in the US and outside the US where that sweatshirt is sold. And I was like, all right, this is pretty cool. Now take that to the article that was recently written, I think it was in TechCrunch or something, where they said, and I definitely could have done this, I didn't, but now I understand how someone would, where instead of me, like I take that picture, but instead of me like searching on the internet, like, uh, You know, Gen Xer. Instead, I just take that picture and I hand it to Agent Mode in ChatGPT and I say, hey, Agent Mode, I saw this cool sweatshirt and I would like to own it. Can you find it for me? And then Agent Mode goes in and finds it and it sends you a message and says, hey, I found it. Would you like me to buy it? And I say yes. And that agent goes to that website, shows up on the website, finds the product, purchases the product. Now, I don't know if you know, but there's a whole bunch of stuff that we do in e-commerce to entice people while they're shopping, right? While they're going through this process. And imagine, if you will, I'm just going to say very liberally, 5 years from now, where everyone is sending their respective agents, right? Hey, ChatGPT, go buy this for me. And it's gonna go, so now the person or thing that we are serving on our e-com sites, if you are selling e-commerce content, like stuff, you're not actually gonna be selling a human with live eyeballs, right? You're gonna be selling to an agent that is basically like shopping on behalf of a user who's not even there. And so what does that, you know, in my mind I'm like, what does that mean? Now, the reason I think this is so interesting is OpenAI has actively pursued contracts with e-commerce providers to give OpenAI a fraction of every purchase provisioned through ChatGPT. So that means that they at least think there's something that's gonna happen in this direction. But I will tell you, I'm already in— I know I'm not your average bear. I am, you know, most of you are probably not gonna go and take a picture of Bobby Flay's sweatshirt and then buy it on the internet. But I feel like certainly our younger generation and certainly, you know, obviously tech-forward people. I don't know. I feel like this could be a thing. But when I saw— when I did, I did that all on my own. And then I saw yesterday this article come out on this relationship OpenAI was creating and the idea that agents are now serving as shoppers instead of humans being shoppers. I just— it also reminds me of like Carvana. Like, I buy my cars from Carvana. Don't judge me. Because I don't like the human experience. Like, I literally hate the human experience associated with car buying. My husband, on the other hand, loves it. Like, loves going in, loves them being like, what car would you like? Let's try this one. Test drive this one. I'm like, no, no, no. I know exactly what I want and just deliver it like cookies or chocolate. Like, just deliver it to my door. So this is my— I have a car right now, bought through Carvana. My next car, If I can help it, will be also bought through Carvana. But the benefit of that is that like, I don't have to talk to anyone. There is no human in that transaction, literally on either side, that I am submitting my request to an AI system, which is their shopper for me. It goes out and finds me a bunch of cars that match what I want. I then submit a bid, and if I get it, then I do have to do the transactions on the other hand. But, but maybe not. I don't know. They might have updated their, their AI processes.. But I can see it. I can see it. Yeah, exactly. Right. I agree with you, Javier, that CTA and conversions. Yeah. I think, I actually think it's a better opportunity because we know how to get AI models like attention. Like, I feel like it's an even more interesting opportunity for sellers to tailor to an AI model. And this is why we, a couple years ago, we came up with this concept of like AI search, or rather than search engine optimization, it was like AI engine optimization. How do you show up in ChatGPT? How do you become top of mind in a platform like that? That, that, that, that in and of itself is a bit of, of work. It's why I encourage you all to do that algorithmic heartbeat that I talked about in one of my classes. Yeah, exactly. No end cap, no last minute things. I don't know. I mean, good, right? Like, that's where I spend I always get the stuff I need the least in the last 10 minutes of my shopping experience. So I think that's fine. But I also think that that whole upsell process tends to be, you know, like that's like the back, the back end of the deal. That's where a lot of money is made for companies. If you know a little bit about McDonald's, right? McDonald's, like they spend billions of dollars to get you to show up at McDonald's and they make 84% of their profit is made on the things you buy after you order, like on your upsells and your additional stuff. It's not made on the burger. The burger basically barely covers costs. It's everything you add to it. The drink, the fries, the Happy Meal. Wait, not the Happy Meal with 4, the Happy Meal with 6, and not the Happy Meal with a toy, the one without a toy. And all of these things have different price points and it, uh, significantly changes. So now that we have an AI system doing this, oh my gosh, it brings me to another piece. Okay. And then I'll, I'm gonna move us to something, um, else to our questions. Um, but another article that came out, who was it? It wasn't McDonald's. I think it might have been, uh, it could have been Popeyes. Anyway, a large, uh, I should look it up, a large franchise, um, food franchise has once again discovered that humans are crazy and weird and we say weird things. And so they launched an AI drive-through and an AI ordering service through the drive-through. And after multiple challenges with humans that say weird things they decided to sunset that project. So you can look it up. And actually, it's funny, if you look up like, uh, uh, food, fast food providers and AI, you'll probably see quite a few examples of this failure. But it's interesting because the thing, the reason that this one, and I, like a few, 6 months ago, it was actually not really fair. Um, because the person who was talking to the drive-thru was intentionally, like, they were hacking it, right? Like, they were intentionally trying to get the model to do something bad. What happened in this case? And so you could use this in your search term because I can't remember the name of the company. If you all remember, definitely type it in. But the person in the drive-thru ordered 18,000 waters. So, and, and it processed, right? Like it processed and ended up inside the store. And, and like, it's, it's like, I guess, equivalent to someone putting a bunch of plaintext in a message, like in an, you know, what do they call it? Denial of service, right? It's like a denial of service attack in the age of AI. You know, like I'm gonna ask for 18,000 waters and 40,000. But the problem with 18,000 waters is that it now is submitting 18,000 water tickets right into the drive-through like area. And obviously waters don't even cost anything. So now the, I mean, it was a bit of a disaster. Oh, what? Thank you. Oh, was it Taco Bell? Oh yeah, I think it was. Awesome. Yeah, yeah, this is it. Oh, cool. See, I remember stuff. Um, cool. Thank you for bringing that up. Awesome. See, if you were in the Zoom, you get access to all this stuff. So it's on BBC. Perfect. Um, but yeah, I just think like this is where as an AI leader, as someone who advises companies, I use this to like, one of the reasons I think this is such an interesting story is number 1, AI red teaming, right? The concept of getting a bunch of different types of people to ask random questions of your model and get it to do weird things. The more broad that team is, the more broad in their understanding, in their diversity, in their lived experience. Someone in there you're gonna hope is gonna be like, well, what if I ask for 18,000 waters? Like, you want that person in your room. You wanna catch that in beta. You might not be able to. But then you should certainly have some guardrails around this. Now we learned this at Amazon Alexa, right? I remember in like the first 6 months of Amazon Alexa, it was super exciting. News reporters would go to people's houses and be like, hey, I heard, you know, Alexa's out. Did you buy one? And when people would say yes, they'd be like, oh my gosh, what was the last thing you did? And they, those people, and one was a child, like a 12-year-old, was speaking to the news reporter and on the news repeated the favorite thing that she asked Alexa, which was, Alexa, buy me cookies. And 500— and there's a reason for this— but 500 orders for cookies went out simultaneously on every— like, up to 500 is when it stopped. But the first 500 Alexas that triggered ordered cookies on behalf of their— and technically it put it in the shopping cart, but still If my Alexa heard that news broadcast, it would've ordered cookies. It doesn't do that anymore, thank God. Um, but that's because we as engineers did not even think that Alexa would respond to a, a news broadcast playing in a different room. We should have thought about it cuz the mic, microphone array and echo cancellation and beamforming is excellent on Alexa. So we should have maybe considered it, but we didn't. And so We then get— now, thankfully, we set a threshold. We said, what are the odds that 500 people will order the same 500 things at the same exact time? So we set a threshold to say, if this happens, it's probably not a good idea, so don't do it. And so it ended up not causing major havoc. But I mean, we should have probably been like, if 5 people do it right, like that threshold in this— I talk about this all the time— the threshold of doing it, not doing it, that is our— us as a human, that's what we do. We decide. We decide if the answer we get back, is that worth submitting a diagnosis to a patient or is that worth considering, right? Is that a definitive yes? Is it a maybe or is it a no? And those thresholds we provide as instructions to a machine. We tell the machine like yes or no anyway. And so this is an example. I, I mean, I guess you could talk about— you could talk to the team and they'd probably say something to the idea of like Well, who would have thought somebody would do that? But you can bet, I mean, at Alexa, we have so many examples of people saying crazy stuff. The two most common things people said in the first two years of Alexa were very nice though. The two most common utterances were, I love you and I want to marry you. So that's awesome. I love that. Great. Yay. But very quickly did they become, you know, just hacker behavior, inappropriate behavior. And that is an example here. And so the other one, I don't— I think it was Popeyes, but it was also in the UK. Okay, a consultant actually was in the drive-thru and was asking intentional questions to be like, to get it to, to mess up. And at the end of the— in my opinion, testing is everything, right? Like, your journey only starts when you have a beta group of testers, right? Like, and you want people to know that they're testing it. So you maybe say, hey, this special location is gonna have an AI model. It's for the next 3 days. Right? You can come and get, I don't know, maybe one free taco. Nobody buys one, right? So one free taco for everybody who comes and tests out our AI drive-through. Like adoption, go-to-market. Come on, people. And then you use the data that you got to customize and optimize your model. Anyway, anyway. So I'm gonna head on over to our school community. But it's interesting, right? We got some cool stuff going on in the world. But I will say we are also repeating the same mistakes. And I, I would love to hear your ideas on that. Like, how do we stop repeating the same mistakes? How do we educate people to know like, this is a problem that we run into, so don't do that anymore. But the reality is, is that when everyone's in like baby tiger mode, right? In their little baby AI and they're building their own model, they think that they are special, that they're building something unique that no one's ever built it before, and they don't even look. For back, like, similar experiences, or they don't even search on the news of big companies that might have failed doing the exact thing that they're trying. So that's why I always encourage us to, to ask questions, ask better questions, um, early, early in the process, because I don't think companies are asking enough. Um, so we had quite a few people join us this week. We are at, let's see here, We are 1,879 members. So welcome everybody. Super excited to have you here. I'm trying to see if I recognize anybody and I don't. So that's exciting. So everybody who got invited, and thank you for those of you who invite people to join. As we launch different programs, we have a $97 AI Solutions Design product that's releasing soon. We have different keynotes that you get to pay to to be a part of. When they, you know, whenever anyone buys something and you were the person that invited them, you get 50% commission added to your school account. Isn't that cool? Like, that's pretty awesome. So if you invite someone to become a member of the I Love AI community or the AI Leadership Institute, you get 50% commission on anything they buy while they're here. And that is my interest in number one, creating good people that want to bring people together. Sharing good content, but also us like bringing more people like us together. So hopefully that will be some incentive for you. But let me go ahead and take a look. I'm gonna go down 4 down and take a look at the question. Alright, that's not a good question. There was no question. So I don't know, I might have to kick that person out now. Dang it. So if you don't ask a good question, you become shady, and I have to like really consider. Okay, cool. This is a good question. So Jacob, one of our new members, he asked a question about deep research, and I'm presuming he means deep research in the sense of reasoning. And so it's a really great opportunity to revisit a feature that was launched recently on ChatGPT that is called Agent Mode. And so Agent Mode is actually the child of two other features. Deep research, as well as operator and agent mode. When you're inside ChatGPT, specifically GPT-5, it allows you to ask the model to go do stuff on your behalf. And that, I think, is pretty cool. Deep research allows you to go beyond a simple web search and actually find and look for— I mean, when you think about, I'm going to do deep research on something, I'm looking for correlations, right? I'm looking for multiple data sources and drawing correlations and insights from those multiple data sources. So it goes beyond, let me just do a Google search, to how do I look in multiple sources from multiple resources and correlate insights across all of them and share them back with you at scale? Because you can ask this upon many, many questions, and the questions can have questions within questions. So this capability is also known as reasoning. I did an entire class class on LinkedIn. I think we— I don't know, I can't remember how many. I have about 3.4 million learners on LinkedIn that have taken content that I've taught there. And in the— it's called— I think it's called Reasoning Engine or something like that. Maybe it's online search. But oh yeah, the future of online search, because this correlates directly to that agents buying stuff for us, right? Like, I feel like the future of online search It started maybe 8 years ago when I could take a picture of something and tell an AI model, hey, find something like this for me. I mean, that's not necessarily new. But adding that to an agentic workflow, adding that to an LLM conversation that says, oh, I was able to find it, but it's actually in rose gold, not real gold. How, you know, do you really need gold? I mean, my husband, for example, has this pair of shoes that he absolutely loves, but they smell terrible. So I'm like, you got to get rid of them. And now, so I went in and it turns out— I was, I was just gonna say I wish I had them so I could show you. I don't wish I had them. But in this, um, shoe, when you flip up the, the tongue of the shoe, it has a QR code. And that QR code sends me to that exact product on the website. But of course the product on the website doesn't— it's not selling anymore. They have a new version. But like Again, I would like to just say, hey, AI model, here's a QR code of a product. I want another one. Go do the magic and have it show up at my door. It is the vision of— do you all remember? Some of you might remember Dash Replenishment Service, DRS, at Amazon, where we had these little buttons. And when you were in the, in the washer and dryer, like the laundry room, you could put a Tide button. And every time you were getting low on Tide, you just press the button and a couple maybe in a couple hours, maybe a day, a new box of Tide or a new bottle of Tide would show up at your door. And they had this for everything. They had it for Tide, they had it for wipes. I was in baby mode at this time. My, my third son was born the, like, 5 days before we launched Alexa. So I was heavy in baby mode at this time. So you could do it for diapers, you could do it for wipes, you could do it for cleaning supplies. Anything that would be— but here's the benefit, right, is that I, I now contextually am going to be drawn to pressing that button. Like, if I'm in the moment, I'm like, oh, I never realized it. And thank goodness, I mean, I am very blessed in, you know, the work that I do that I can afford to anytime I need to press a button and have whatever I need show up at my door. Like, what am I— what world am I living in? That is not my parents' world, right? Um, where we were like, hey, you ran out, you're gonna have to wait till next payday kind of thing. So So, but that is, it's an interesting thing to think about because as we start using AI, I've heard these terms. I'm just gonna throw this out there to like fruit basket turnover our whole conversation. But, you know, universal basic income, right? Like as AI starts doing its thing and starts impacting workforces and I don't know, like maybe everyone needs an, you know, a universal basic income so they can press whatever they need and have whatever they need show up whenever they need it. But I don't know, we'll see. I, we're, we're, I think, fast approaching a time where those considerations need to be deeply, deeply considered. They are actually in some rooms right now. So, all right, let me go to the next one. So reasoning, yes, you could go look at it. Hopefully I gave you a good description of what it was. And you can always ask deeper questions in the community itself. Oh, good. All right. Well, the next one was they just wanted to understand basic terminology. And so what I will do Now in our last few minutes, I, of course, I think I mentioned this on the last call, but I wrote a book. It's called Scaling Responsible AI. I actually am just now confirmed to be able to maybe bring you another one. So this Scaling Responsible AI book, it's on Amazon. Of course, it's everywhere books can be sold. Barnes Noble, my favorite place. I always tell people I went to the University of Barnes Noble. That's how I learned to code. But what I like about this book is that they're like these types of questions, reasoning, online search, you know, for example, terms and definitions and just understanding the lay of the land. This book will certainly be helpful to you. You can get it in Audible format. I have a wonderful, beautiful person. I never met her. I just know her voice, but it sounds pretty. And she, she was the narrator for my, my book and it was done beautifully. So thank you to her, by the way. But I'm just going to flip to any page and kind of give you an example of what I mean by content. And we'll end with this, and this will also be what we talk about in our breakout sessions. So every week, my favorite part of our conversation is when we get to know each other. And so right now I'm going to read the section I just happened upon, and then I would like us to take this into our breakout rooms and just talk about what this means to us. Some of us are dealing with whatever this is right now. Some of us want to deal with it in the future. Some of us, maybe know better than to deal with it at all. I hope that's not the case for this, but in this case, I happen to land on building a resilient AI team, which I'm a big fan of. Building a resilient AI team is essential for navigating the ever-changing AI landscape in this digital age. I'm hoping I don't see any words that the LLMs, right, say that they say all the time. I hope not. I definitely didn't use AI for this. It was before its time. Anyway, in this digital age, the success of any AI initiative hinges not just on cutting-edge technology, but also on the strength and adaptability of the team behind it. A resilient AI team is one that can weather challenges, pivot when necessary, and continuously learn and grow in the face of adversity. So how do you go about building such a team? Well, it begins with fostering a culture of collaboration and diversity. Now, I say that, and I'm— I don't know if you all saw. Anyway, Sky knows. But somebody said some things about collaboration and diversity, like, literally those words that are in my book. But this comes from me being at Amazon. And when I was at Amazon, I was a TPM. I managed a bunch of engineers that were building the models. I didn't build the models. I was not a data scientist. I'm not classically trained as a data scientist. I call myself, I guess, a— what do they say? A citizen data scientist, right? I learned enough data science so that I can look at a model and go, that's not cool. You should not do that. We should fix that. But I wasn't classically trained at Stanford or MIT or any of these places. Like, I learned through a book. On my own. So that being said, but as a TPM, I know how to manage engineering work to get work done and get stuff into production. And what I realized is that the quality of your team is directly correlated to the quality of the models that you build. And what I mean by that is that I always say you need a symphony of talent, like all sorts of people, because those people are gonna bring up ideas. And, and I would, when I was at Amazon, we would have meetings And it was funny. I remember being in a meeting and we were in like a WeWork, I think, and there were like glass doors. I don't know if you've ever been to a WeWork with glass doors, but they were like glass doors. And so when we were in a meeting, people would walk by and after the meeting they'd come up to me and be like, is everything okay? Like, are you okay? And I'm like, yeah, why? They're like, I mean, it looked like it got kind of heated in there. And like, I'm like, oh no, no, that's just like diversity in action. Like everybody gets to say their piece. Some people get very— but at the end we're all like, cool, bro, see you later. Like, we're family, right? Just like family. You know, you have an argument with a brother or sibling and you're like, ah, and then 5 seconds later you're like, hey, you want to go get some pizza? Yeah, let's go. Right? That is the kind of culture you need. You need people to be brave enough to stand up for what they believe in. But the resilience part is a principle that I learned at Amazon that I still use to this day, this concept of disagree and commit. That, yeah, you might have a perspective and you might have, you know, a really good reason. But at the end of the day, the leader has to decide what the, the correct way is to move forward. And if you are not the leader, which I encourage you all to become, it's a way better place to be. But if you're not the leader, if you are working on a team and a leader is making those decisions, when they make a decision that's not in alignment with what you think is the way to go, and I don't mean ethics, I don't mean like, hey, I don't think we should kill people and they say you should kill people and you go like— I'm not talking about jumping off a cliff because your friends are jumping off cliffs. I'm talking about like, just, you know, you know from your lived experience that this project should go in a different direction and they just choose not to go that way. Disagree and commit means that you're going to commit to the new path after saying your piece, after being heard, after being documented that you said your thing. You're then gonna commit to the new path, the new idea that your leader is moving forward with. You're gonna commit to it as if it was your own idea. This is that resiliency I'm talking about is not getting stuck in, like, I know it's not gonna work. It's never gonna work. It certainly won't work, by the way, if you have that attitude. But, but, like, releasing, you know, being willing to, like, say your say, but then in an AI project, releasing those very specific things that you you know, wish were gonna be done differently, adopting the new plan and not holding a pocket veto in your pocket waiting to say, I told you so. This has been personally very hard for me because I know things and I worked on stuff and not everybody has. And so disagree and commit is very personal to me because I literally need to be very careful about, okay, we're not going with my idea, which is fine, but I need to commit to the new path like as if it was my own. Like with that level of integrity and that level of intention. And when you have a team that does that and does it like with a kind heart, if you will, you will find that even when things break down and it comes back and I'm like, y'all remember when I told you about that? Like, it's lighthearted. It's not resentful. It's not mean. It's like, oh man, yeah, well, we tried. And then of course, I get my day of like, I don't have to say I told you so, but y'all know I told you so. But that disagree and commit is a really interesting philosophy. So I'd like to take the, the next few minutes that we have together to just talk about what resiliency means to you all as a member of a team and specifically how it impacts AI projects. Like, I feel like a lot of, a lot of AI projects, they often happen due to the sheer persistence of a team to continue to work. OpenAI took 7 years to get to market. It. Like, the idea took 7 years of, it's not working, it's not working, it's broken. It's like, that team had to be willing to come into the office every day and be like, it's still broken, it's still broken. Like, that is resilience. Like, how does that show up in your world? Um, maybe what are the characteristics of people like that? Uh, do you share some of those characteristics? I just think it's a good, good topic of conversation is like, what does it take to build a team that will create safe, responsible, and profitable AI. So with that, I'm going to head off— head you off into breakout rooms. You all have a chance to chat it up. If you can't make the breakout room, no sweat. It's fine. Most of us enjoy it, like hanging out with one person. So if you can't make it, don't worry. Just don't click on the button if you are gonna head out instead. But then when we come back, we'll close out. And if you have any questions, the rest of our time will be dedicated to at answering questions that you have in general. So I'm going to break us out into, let's say, we're going to say 3 or 4. There's a few note takers here. So let me first— oh, good. That one looks good. Let me move you to Room 7 and move you to Room 7. I have note takers, lots of note takers. Room 7. Room 7 is the lucky room for note takers. Okay, let me move you. To room 3. I'm almost there, everyone. Hang in there. I'm moving you to room 4. Thank you for letting me know. Okay, cool. 1, 2, 3. 1, 2, 3. 1, 2, 3. I think we're good. Oh, almost. What room is that? Room 3. All right, let me make sure. I think I've got all the note takers out. All right, so go have fun. Talk about AI resiliency in Teams and I'll join one of you. I think there's one room that has one less person, so I'll join you guys over there. All right. Talk to you soon. I'll be back in 10 minutes. Have fun. Just say yes, click the button, have a conversation, join the conversation. Darrin, are you here? You could hang out with me if you want. Yeah, uh, uh, let's see who else was asking that question. Um, Ricardo. Did Ricardo already go away? He did. What's up, Nitin? Good to see you. We'll have our own little private group, although the team will miss you if you don't go there. Um, let's see. Let me see. Oh, Eve. Oh, yeah, of course. Yes, my dad was in the hospital, but he's better now. So, hey, all right, this is the time where I tell we don't need to keep our live audience sitting and waiting for 10 minutes. But this is where the fun is at, right? Like, if you are on LinkedIn or on Facebook and you're like, I want to be in a Zoom call, I want to be in a room where we're talking about team resiliency and AI, like, that sounds like fun. Head on over to noelle.ai, join our community. It's free, it's fun. We do live calls every week. We have lots and lots of amazing leaders providing really great insight. It's cool. I hope you'll be friends with us. We would love to have you. If you're looking for a tribe, this is the one. You're welcome here. We try to create a safe space for everybody to ask questions about AI that they may or may not be able to ask in their day job. So, so thank you. Thank you for being here. I'm gonna stop the live and thanks for, for joining.