EP 495: Gemini 2.5 Pro Unlocked: Exploring everyday use cases

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. All right, so if you haven't heard Google's new large language model update in Gemini 2.5 Pro, is good. It's like really good, as in benchmarks, the best human preference, the best. But what can it actually do for your business? I think this is something that we're always thinking about. I think early on, you know, in the ChatGPT days, we, we got into this kind of rut, right, when large language models first came out, and we thought, okay, well, They're just for creating content, right? This is to help me write a blog post, or a large language model is to help me, you know, write something for LinkedIn or improve an email to a colleague. Yes, large language models are good for these things. But what about when we talk about state-of-the-art multimodal large language models like Gemini's, sorry, like Google's new Gemini 2.5 Pro? So today I thought we'd have a little bit of fun and maybe a little bit of chaos as we go over in part 2, Gemini 2.5 Pro unlocked, exploring everyday use cases. All right. I'm excited for this one. I hope you are too. What's going on, y'all? My name is Jordan Wilson. If you're new here, thank you for joining us. This is Everyday AI. This is your daily livestream podcast and free daily newsletter helping us all not just keep up with AI, but how we can all actually use it to get ahead, to grow our companies and to grow our careers. Is that personal? Is that you? Is that what you're trying to do? Uh, if so, this is step 1, listening to this podcast or livestream. Uh, step 2 is going to our website at youreverydayai.com. Here's what we do. 2 main things in our free daily newsletter. One is we recap and sometimes summarize, uh, you, you know, the, uh, the episode for today. You know, sometimes I have guests. On today. It's just me talking about Gemini 2.5. So we give you what you really need to know and pull the valuable insights from each day's episode, as well as keeping you up to date with everything else happening in the world of AI. So make sure you go to our website, youreverydayai.com, sign up for the free daily newsletter there. All right. So normally we go over the AI news and all that in the beginning of the livestream. This one could be a longer one, and I'm trying not to. So if you do want the AI news, we're going to have that in the newsletter. All right. So I'm excited and I hope I can get a little bit of help from our livestream audience today. So thank you for tuning in. Dennis joining us from New York City. Yeah. Where are you all from? I should ask this more, right? I like to, I like to know where our livestream audience is from. Brian joining us from Minnesota. Kyle, thanks for tuning in. Michelle, Big Bogey, Sandra, Jay, everyone else. Thank you. But I might, I might be asking some, some help from you all today. All right. But let's just get caught up. Right. So I did an entire episode yesterday on what's new in Google Gemini 2.5. So if you do want to know, just scroll back one episode. Maybe if you're listening to this on the podcast, it's episode 494, where we just went over the basics of Google Gemini 2.5. But as the world's fastest recap here, here's kind of the super simplified version of what's new, okay, in Gemini 2.5. So it has built-in thinking. That's the biggest one. It is technically a hybrid model. It combines the kind of old-school, quote unquote, transformer model with a reasoning or chain-of-thought model. So you'll see that as we do some live demos here. And it's gotten very impressive scores, not just on traditional benchmarks, but also some newer benchmarks like humans' last exam, humanity's last exam, where it scored way better than any other large language model. It does have an enormous 1 million token context window. So that is more than 1,500 pages, as an example, 30,000 lines of code before Google Gemini 2.5 Pro begins to forget things. I will let you know, and we'll probably see here live, that is when you are using it in AI Studio, versus the front end of Google Gemini. More on that in a minute. Probably one of the biggest leaps in terms of capabilities, and maybe this will be applicable for your business, maybe not, is the advanced coding. So Gemini 2.5 is very, very good at coding. All right. And you might be thinking, all right, Jordan, that's not me. I'm not a software engineer. Right. Okay. If you listen to our 2025 AI roadmap and prediction series, I said in 2025, Everyday non-technical people are going to be using large language models to spin up their own apps, to spin up their own, you know, I don't know, Chrome extensions, their own desktop apps that help them do things better. We're not there yet, but I think we will be there very soon. So keep that in mind. You know, just because you're not a current coder or developer or software engineer, you should still, I think, really pay attention to this Google Gemini 2.5, the big leap in coding. And maybe some of our use case examples will show that. you know, number 1 benchmark ranking, that's huge. So the biggest thing we talk about, there's all these benchmarks that I think sometimes, you know, AI labs overfit for. But when it comes to ELO scores inside the chat, the LM chatbot arena, that's human preference, right? So people put in all kinds of prompts, you know, write a blog post, create, you know, write code for this, you know, generate a creative outline for, you know, or strategy for X. right? Um, and you get 2 responses. You don't know who they are. You choose which one is the winner. Uh, and Gemini 2.5 Pro has literally broken the record for the biggest leap into the number 1 spot. Normally when a new model comes out, you know, from OpenAI, from Claude, et cetera, it'll usually get the number 1 spot, right? Because generally it is, you know, between 2 to 6 months, uh, between big models, especially earlier in 2024. Uh, so, you know, usually the, the top model would come in a couple points higher. Couple preference points higher. Gemini came in at 39 points higher, uh, than, uh, what was in second place. Now what's in second place right now is GPT-4o. Uh, the other, uh, big recap here: it's free. Was not expecting this. So Google did not even announce this in their initial Gemini 2.5 announcement. They quietly put it out in a tweet over the weekend, uh, right? But even if you don't have a paid account of Google Gemini, you do have access to Gemini 2.5 Pro for free. The limits are a little more restrictive. All right. So, uh, one more thing, one or two more things before we get started here. So, uh, and, and, and livestream audience, you know, if you have any things you wanna try live, let me know. Maybe, uh, I don't know, in your comment, uh, I should have thought about this, but like beforehand, I don't know, maybe put two stars. All right. And then I'll see if I can maybe copy and paste it. Uh, I don't know if I'm able to, but I'll try, or at least I can try to, you know, get the gist of, of what you wanna see. Uh, but before we get started, a couple things to keep in mind. Our podcast audience, thank you for tuning in. Y'all are awesome. Uh, I never would've thought when I started this thing, this would be a top 10 tech podcast, but this is one of those. You might wanna check out the newsletter so you can come and watch the video. You can always rewatch it on our website, uh, on YouTube, on LinkedIn. I'm gonna try my best. to verbally describe what's going on with this is unfortunately gonna be a very verbal, or, or sorry, a very visual episode. Uh, and this is something, uh, this is always the number one request we get, right? Do more live demos, do more live demos. Uh, so, you know, podcast audience, uh, I'm gonna try my best, but this is one you might wanna come watch the video on. Uh, another thing to keep in mind, AI Studio versus Gemini. Okay? Gemini is the front-end chatbot for Google. AI Studio is kind of a sandbox for developers, although it's not as hard as you may think. Right. You know, there's some initial setup, but then after that, it's pretty easy. You know, if you're on a paid plan of the front-end Gemini chatbot, you can turn off model training, which is important, right? Because you should never be sharing, you know, proprietary, sensitive PHI, right? Like private health information into a chatbot. But if you are using AI Studio, there's no turning off data training. So AI Studio is free. That is actually where you get the more powerful version of Gemini 2.5 because you get the, the entire context window and some other controls that you don't get on the front end of the Gemini chatbot. And hopefully I'll be able to demo that here in a minute. But just keep in mind, Google's AI Studio is free, but you cannot turn off data training. If you are on a paid plan of Google Gemini on the front end on the chatbot, You can turn off, um, training. All right. The other thing I'm doing this live, y'all. All right. So bear with me. Uh, but I think it's actually important, right? Because if you go watch anything online, you know, there's some great creators out there, um, you know, who put together, you know, demo videos and all that. I know a lot of these people. I talk to them and I know how long these videos take, right? So, uh, sometimes to, to put together a couple demo use cases, of something like Gemini 2.5, it might take them 5 hours of recording, uh, for a 20-minute video. Okay. And a lot of editing to make sure it looks right. I don't like that. You know, people are always roasting, uh, roasting me on our YouTube channel because it's like, oh, your production quality stinks and you have all these mistakes and sometimes you stutter or say the wrong word. I'm a human, right? This is live. This is unscripted. This is unedited. Uh, this is just, you know, but I think it's important because I think so much of these, uh, of these demos of all large language models that you see, all AI tools are overly polished. They're manufactured. You know, in some cases, you know, they're being artificially pumped and promoted on the backend to make you think there's something that they're not. This is real. This is live. This is unedited. All right. So keep that in mind. Live demos with generative AI are a terrible idea, right? But you all like them. You all want to see them. So we're going to do them. And so far, my takes right now with Google Gemini, it has an extremely high ceiling, but a finicky floor. All right. Let me, let me kind of describe, uh, what I mean by that. So here's, here's an example. And, you know, I put this out on Twitter and I'm going to ask the Google team about this, but, uh, it's, it's keep in mind Gemini 2.5 Pro is experimental. All right. Very experimental because sometimes you're going to get a weird, uh, a weird result like this, right? I always have a series of prompts. That I use to, especially for internet-connected models, so I can make sure that they're correctly pulling information, right? When we talk about the role of human in the loop, it's very important. And as large language models get more powerful, more robust, more features like Gemini 2.5, I think us humans think, oh, we can sit back and relax. We actually have to be more vigilant. The more that we hand off to large language models, the more that we have to— I like to think of it as expertise in the loop. right? Not human in the loop. Human in the loop just thinks like, okay, you know, I'm gonna blindly, you know, do my human job here. This looks good. Click. No, you have to apply your expertise. This isn't a simple example, right? But I said, what's the latest episode of the Everyday AI Show by Jordan Wilson, right? I want to see if Google Gemini 2.5 can get my episode from yesterday, right? And in this example, you know, because it is a hybrid model, I can even see the thinking. And it says the user is asking for weather forecasts in Chicago, Illinois for today. April 1st, 2025. I should use a weather tool to get the current weather and forecast for Chicago. Number 1, not true, right? It didn't. Number 2, not surprisingly, right? It picked up my location without me telling it. All right, so keep that in mind. It's finicky, it's experimental, but when it works, I am very impressed. I am very impressed. All right, let's get wild. Let's get wild, y'all. Please, livestream audience, can someone tell me if you can see, see the screen here. I'm gonna be jumping between some, some tabs here. But if you could let me know, 'cause I don't wanna do another 25 minutes of the show and bringing you guys these live demos and you're like, oh, Jordan, you weren't sharing your screen at all. Kimberly says, we need to see more bloopers too. It's a part of life. Yeah. I think that's how you learn generative AI. That's how you get better at large language models. You, you try them. No one's an expert, right? Or I won't say no one. There's very few people that have been working in large language models since, you know, for 10 years. There's a couple people, right? But most of us, you know, you have to learn on the fly and you learn by failing and you learn by making it better. All right. Dennis, thanks, Dennis. Dennis said AI is cool, but we love humans more, Jordan. Okay, cool. All right. Thank you, Nicole and Kimberly for letting me know and Charles that you can see the screen. Cool. Let's, let's, let's get after it. All right. I'm gonna be jumping around a little bit here, y'all. And, uh, I apologize if you hear like a lot of clicking. All right. Uh, that's my mouse. Uh, I should probably figure out how to, you know, not pick that up in the podcast. All right. So I'm gonna go in and, uh, upload a file. So first I am right now, I'm on the front end of Google Gemini. Uh, in, in your dropdown, you have 2.5. One thing to keep in mind, And maybe this is a hack for ChatGPT. There's no model switching, which I wish there was in Google Gemini on the front end. So as an example, if I start in 2.0 Flash, you know, I'm just going to say sup. All right. Now, if I want to model switch or start working in 2.5, I can't. It refreshes that chat. So why? Why does that matter? Why is it important? Well, as an example, I'd love to, like, use Deep Research. so Deep Research inside Google Gemini has been upgraded to, Gemini 2.0. It's actually amazingly good. Uh, but so if I wanted to, you know, do something in Deep Research and then go over to 2.5 Pro, you can't do that. Whereas with ChatGPT, you can. I think that's like such an underrated hack is just model switching, uh, inside ChatGPT. But, you know, before we get started, uh, it's worth pointing out. All right. So, uh, I am on the gemini.google.com. I have a paid account, FYI, but Even if you have a free account, uh, you should be able to do this. Livestream audience, if you wanna follow along, you know, you can do that as well. All right, so I'm, uh, selecting 2.5 Pro Experimental, uh, from the dropdown menu, and I'm going to add a file here. All right, so I'm going to add a PDF here. AI moves too fast to follow, but you're expected to keep up. Otherwise, your career or company might lag behind while AI-native competitors leap ahead. But you don't have 10 hours a day to understand it all. That's what I do for you. But after 700+ episodes of Everyday AI, the most common questions I get is, where do I start? That's why we created the Start Here series, an ongoing podcast series of more than a dozen episodes you can listen to in order. It covers the AI basics for beginners and sharpens the skills of AI champions pushing their companies forward. In the ongoing series, we explain complex trends in simple language that you can turn into action. There's 3 ways to jump in. Number 1, go scroll back to the first one in episode 691. Number 2, tap the link in your show notes at any time for the Start Here series, or you can just go to starthiereseries.com, which also gives you free access to our inner circle community where you can connect with other business leaders doing the same. The Start Here series will slow down the pace of AI so you can get ahead. All right. So I'm going to describe what's going on as this happens. I'm going to say, please. So I'm just saying I'm uploading a PDF and I'm saying, please transcribe every word of this. So this is about a— let me see how many pages this is. It's probably about a 15-page PDF here. So these are, you know, people reach out and they're like, hey, I wanna, you know, advertise on the Everyday AI podcast. So I have this, this little deck that I send potential advertisers sometimes. So hey, if you do wanna reach one of the largest audiences in artificial intelligence, you know, on our podcast, you know, make sure to reach out to me. But the thing is, most large language models cannot read this. Because I mean, number one, I made it in Canva. So, you know, most, most large language models when they're using computer vision, when they're using sometimes OCR technology, they all work a little bit different. They really struggle with this because it's all essentially images, right? It's not like I build this in Word and it's a bunch of text. This is very visual, right? There's backgrounds, there's tons of images on each page, right? It's, it's a lot going on. So, you know, even to pull all of these words I mean, we'll see. I've done some of these so far, some I haven't. So let's see how Gemini 2.5 does. So I can click show thinking, right? And I'm not going to be able to do this for every single one. But it says I need to get the relevant content to answer each user's questions. The user wants a transcription of the entire PDF document. I have the extracted text from the document provided by the content fetcher tool. So Why I'm going to spend a little bit more time on looking at the chain of thought. And y'all, this is huge, right? The thing I love about Google Gemini's chain of thought is you can see their tool usage. All right. Which is going to help you get more out of the tool if you know, because you can start to speak Google's language. And hopefully that will, you know, become a little more clear here when I try another prompt here. So anyways, let me just go ahead and scroll down and you'll see right away it's breaking it down. Page 1, here we go. Everyday AI sponsorship opportunities, daily podcast, livestream, newsletter. Perfect. It's got the website. Great. Page 2, it's got it all. Okay. This is really, really good. I haven't seen this, uh, out of a large language model yet. And it's, it's formatted. It fixes, you know, sometimes the, the, the fonts look a little weird. Um, you, you know, but it crushed it. All right. This is, this is impressive, y'all. All right. So, uh, I'm going back. So at the bottom I have trusted by leaders from, right? Because we have all these people from big companies that have, you know, that read our email newsletter, that reach out to me, that have given us testimonials, you know, from Google, Amazon, NVIDIA, Microsoft, et cetera. Right? Uh, we have a lot of listeners. Yeah. If you want to reach them. Uh, so not only did it get the text, but Google Gemini here, very impressive. Used computer vision and gave me just the names, right? I didn't put the name Google, the name NVIDIA, the name IBM. Those were multiple images. Mind-blowingly impressive. All right, page 3, you know, partnership opportunities. So good. So good. So I'm actually curious. And again, I'm doing a lot of this live. Did you guys know, hey, livestream audience, did you guys know this? I, I, I didn't even know that it was gonna look at the images in this deck. Uh, you know, I've tried this a lot with ChatGPT. I've tried it a lot with Claude. Um, I haven't tried it with the updated version of 4.0 that was just rolled out a couple of days ago, so maybe it'll do better. This is very impressive, right? Uh, so I'm curious if it's even gonna pull, uh, some of these stats. So I have all like our ad channel overview. And there's like text within screenshots of this image. So, you know, I'm, I'm curious. I'm just gonna scroll down to, to that page. Let's see here. Ad channel overview. Okay. It didn't pull it in, but that's fine. The text was probably too small, but it literally crushed it. My gosh. It even created, I have a chart. This is so, so good. I have a chart and it converted my little chart, which is just, I made in Canva, right? So not only, wasn't able to pull all that because a lot of it is images. It created a chart for me that I can export to Sheets so I can click that export to Sheets and then open in Sheets. Bam, there it all is, our little breakdown. Just that right there is wild, y'all. How many times when we talk about business use cases, right? I don't know about you guys. I read a lot of PDFs, right? Or a lot of documents. Sometimes you may not have the version that you need, right? It's like, oh my gosh, this was from Bill. He left 2 years ago. I have to redo this entire thing. Well, you can upload it into, uh, Google Gemini 2.5 Pro. It's going to transcribe the whole thing. If there's charts and graphs in there, it's gonna recreate them. You can open them, uh, in Google Sheets. This one use case alone. Wow. Wow. Very, very, very good. Uh, all right. Hey, cool. Sandra says she's doing it along on her computer. All right, let's do another one. And this is where I think we're going to get some things that go wrong, but let's try it anyways. All right. Because like I said, I did try some of these. Some of them I did not. So I'm saying, find the 20 latest episodes of the Everyday AI podcast and give me a brief summary of each one. Then find 5 trends between episodes. All right. So think. What's your business use case? What are you following and think? You know, obviously Google Gemini 2.5 is connected to Google. So one of the reasons I'm doing this, I think it's going to fail. All right, here we go. Hey, we got a live hallucination, y'all. All right. So it says the user is asking for the date of Easter in 2025. Strangely enough, this is the exact same hallucination I got the first time I tried it. So I'm just going to add one more thing. I'm going to put my name by Jordan Wilson. I don't think so. Last night I did get this to work correctly. But I did get some interesting, some interesting insights by looking at the chain of thought, by looking at the different tools that Google is using under the hood to pull this information. All right. So now on the second time, it got it right. It didn't tell me the dates of Easter, which I don't know why I did it. All right. So it's breaking this down. So it says this requires multiple steps. One. Oh, It just shrunk that. Okay. Did you guys see that live? It was working correctly. Everything was good. And then it says the user is asking for the top 5 rock songs released in 1977. Y'all, this is why I said, I said this ahead of time. The ceiling is so high, the floor so finicky. Uh, at least right now on the, uh, front end of Gemini 2.5. Pro. So what we could do, I wasn't planning on doing this, but let's just do it anyways, y'all. Let's go into AI Studio. All right. So AI Studio, it is more of a developer tool or a sandbox, but it's actually very easy once you get it set up. All right. So you can click the create prompt button right here. You can choose the different models over on the right-hand side. So a little different. I'm going to try the same thing. Let's go to Gemini 2.5. Pro Experimental. I'm going to turn the temperature down on this. Okay, the default is 1 for creativity. I want facts. All right, and then I'm going to go ahead and turn on— so you can turn on and off different features. This isn't a full-blown AI Studio tutorial. I just want to see if this will work. All right, but I'm turning on grounding with Google Search. So I have found when I get some weird, uh, little hallucinations like you just saw on the front end of Google Gemini, when I usually, when I try it inside AI Studio, it works a little better. All right. So now I can expand to see the chain of thought. So it's saying the user wants a list of the 20 latest episodes of the Everyday AI podcast. Identify 5 trends. So it's looking up search queries. These are the search queries. What are the latest episodes of the Everyday AI podcast? Everyday AI podcast latest episodes list. Right. It developed a plan and then it says, here are the 20 latest episodes of the Everyday AI podcast. All right. I spoke too soon. I did not think Google Gemini was going to get this correct. We saw when we used the front-end Google Gemini chatbot, it went off the rails. It's experimental, y'all. It's, it's, it's, it's going to do that, right? But inside Google AI Studio, very good job. So interestingly enough, it got this 100% right. So we got our latest episode, which was 494 from less than 24 hours ago. so it did a good job. Uh, and then it got the most recent 20. Fantastic. Uh, now I'm— now it says 5 trends between episodes. Uh, so it says, uh, there's been a focus on major AI players and models. Correct. Uh, rise of AI agents and automation. Yep. Uh, industry-specific AI applications, impact on work and productivity, hardware and infrastructure importance. Great. So it did a good job of picking up, you know, some kind of some common trends, uh, over the last 20 episodes. So, uh, even though Google Gemini, uh, the chatbot got a big fat failure, the Google AI Studio. Very impressive job. I've done similar prompts like that between all the internet-connected large language models about 6 months ago, and none of them handled them the way that Google's AI Studio just did. All right, let's try another prompt here. Here's what we're doing. This one's a little tricky. All right, I'm saying summarize this page, and I'm giving it a Boolean search URL. All right, I'll explain what that is, but the reason I wanna do this is to look at the tool use, right? So look at the chain of thought. So you can click show thinking when you're using Google Gemini 2.5, and it says the user wants me to summarize the content of the Google search results page. And then it says the browse tool can be used to extract information from a specific webpage URL. However, the URL provided is a Google search results page. The Browse tool description explicitly states not to use it for Google search result URL. So instead it's saying I can use the Google search tool. And this is a huge— I'm not going to say cheat code, but this is going to save you so much time once, you know, this Gemini 2.5 Pro on the front end becomes a little bit more stable. Because now by looking at the chain of thought, you will know what exact tool that you need to call because Google doesn't necessarily tell you. right? So just in case you're curious, this, this Boolean URL, it's essentially, like a Boolean search operator, that I use. I do this every day when I go and see what's the most important AI news, right? But it's just search results for certain companies, you know, OpenAI, Apple, NVIDIA, Microsoft, Amazon, Anthropic, et cetera, the latest news. So it's the last, the last 24 hours, just AI news from those companies. so let's see, what Google Gemini ultimately did. So I just said essentially summarize it. Did a good job. Did a good job. So it says key trends, major players are rapidly releasing enhanced AI models like Google Gemini 2.5, OpenAI's, GPT-4.5, Anthropic Claude 3.7, IBM's Granite 3.2. It did a really good job, right? Uh, even though I can't see exactly, oh, did it go to all of these pages? Did it just look look at the headline and the meta description, it did a really good job. So think business use case. I love Boolean search terms or, you know, Boolean operators, right? Do that for a Google search for what you care about, right? Maybe it's market research, maybe it's logistics, right? Put in your competitor names, whatever. I think there's so much utility for using just Boolean search and AI tools to quickly get you caught up. That on, on things instantly that would normally take a very long time. All right, let's keep this train moving. Choo choo. All right, here's one I really wanted to do, but we're not going to have time. All right, so I'll move on to another one here. All right, let's do this one. I'm saying so for this one, I'm going to use Canvas. So this is another kind of update to the update. So Google Gemini 2.5 Pro was just released less than a week ago, and then over the weekend, Google did a lot of other updates to Gemini 2.5 Pro. Number one, they said, all right, it's, it's free for everyone. Number two, they rolled out Canvas just about a day ago. So Canvas, it's kind of similar. I actually think it brings the best, the best of both worlds between OpenAI's Canvas, which is more of like an interactive, you know, document editor that can render some code along with Claude's Artifacts feature, which can render just like any programming language. All right. So in this instance, I'm saying I'm enabling Canvas and I'm saying create an HTML clone of Wikipedia, but give it heavy Chicago vibes. Make it fully featured, including clickable links and multiple pages that work. Make sure to include the most important Chicago things. Right. I'm trying to have a little fun here, y'all. So let's see if this works. So first it is writing the code. So like I said, it's great at coding. All right. Fantastic. All right. So once it's done, which I don't think it should take very long, it's— there's a preview tab as well. So when I— in— when I start this canvas mode, it kind of takes up the full screen, but I can minimize it if I want. I'm going to Pull this over a little bit so I can see. All right. It should be done here pretty quickly as I take a sip on the coffee and I'm scrolling through the, the, the livestream comments here, y'all. I'm going to see if there's any questions. All right. Josh said, look what I created this morning. Go, go check out what Josh created. Um, Charles says, why don't you use ChatGPT for news? I do. I do as well. Uh, so that same URL. I did a whole entire show on how I did this, Charles, using ChatGPT tasks. Monica says, what do you think are some of the best business use cases for this model? I still have a couple here, Monica, but I think one of the best ones, working with PDFs, right? This has been getting accurate information extracted from PDFs and then being able to use that as a baseline, right? Because now I have all that text and maybe I'll do something with it that I extracted from a PDF. That's a, a simple no-brainer. Everyone's working with PDFs and, you know, essentially extracting any information from a PDF if you need to recreate it, if you need to grab some information from there and use that as a start for, you know, creating content, right? So, uh, in my example, I had, you know, our Everyday AI kind of, uh, sponsorship kit. I could then use that, copy and paste some of that information, go into deep research and say, hey, are these, uh, rates accurate according to 2025 popular podcasts or something like that? So, That's, that's one small thing. One small thing I can do. All right, let's look at this. I'm gonna zoom out. All right, here we go. So we have our, let's, let's see, how can I do this? Full screen here. I had this last night. I thought I could. All right. Anyways, we have our Chicago Wikipedia, literally. One shot. All right. So it says, welcome to Shikepedia, your go-to source for all things Chicago from a Chicagoan's point of view. Forget the encyclopedia. This is where the real info's at. So this is a fully functioning Wikipedia clone, right? I can click. Oh my gosh, it works. There's multiple pages on here. It's interlinked. So I can click deep dish pizza, right? And then I can, you know, at the bottom it says all See also Chicago hot dog. I can click Chicago hot dog. The Chicago hot dog, also known as Chicago red hot, is a culinary masterpiece in a bun, right? No ketchup, all beef, right? This is so good. This is so good. It literally created a very small version of Wikipedia, but Chicago style. And then the good thing is I can go in, I can go in and change anything with natural language, right? And I can just say, you know, make it, make it way more Chicago, uh, and more '90s Bulls references, right? Whatever. All right. Uh, so we're going to come back to that one here in a second, uh, and go on to our next use case. That one was fun. What'd you guys think? Pretty impressive, I thought. Uh, all right, let's do this next one here. Okay, here we go. All right. I might not even have time, to read this because it's a little bit of a longer prompt. But I am essentially saying, you know, you're an analytics and research expert using Gemini 2.5. Analyze the sentiment of online mentions of Apple over the past 30 days. And I'm giving it kind of step-by-step instructions. You know, I'm saying essentially look at all of the information on the, on the open web that people are talking about Apple, right? Then identify 5 recurring themes or issues based on sentiment analysis, right? Provide actionable recommendations for Apple's PR team to address any negative sentiment. And then ultimately, I'm using Canvas for this. And then I say, create an interactive dashboard that displays your findings. Make sure to go into insane detail ensuring accuracy and depth. All right. So I actually did this one previously in my first version. Okay, let's see if it does. Okay, look at this. Gemini was, was a step ahead of me, y'all. It actually created two different canvas files within the same, within the same kind of response here. So Okay, so it's building our sentiment dashboard. Cool. All right. So first, here's the sentiment analysis over the past 30 days. So I want to, again, human in the loop, look for accuracy. This is correct, right? It says AI strategy execution concerns. All right. So this is good. It gave us a good text-based report. It gave us actionable recommendations for Apple PR based on real-time, up-to-date information. It gave us 5 recurring themes. Right. Vision Pro's lackluster reception. Oh, weird. If only someone would have told you that 6 months before it came out. Oh, wait, I did. All right. So it gave us a great text document in Canvas. So one thing you'll notice about the Canvas, if you haven't used it, it does have some of those great ChatGPT UI/UX features where you can just change the length, you can change the tone, you can suggest edits. So I can just type live, right? so it's literally like a Google Doc, which is very impressive, right? Even just the Canvas integration from the text-based perspective, is extremely useful for any business use case because right away I can export this to Docs or I can just continue, to type and work with it here. But it created two different canvases for me. Let's see how the other one turned out. Bam. Love it. Uh, it actually turned out not as good as my first. I did demo this one first. but it gave me a very nice looking, kind of interactive dashboard, you know, nice colors. It says overall sentiment mixed slash cautious. It says while investor metrics, you know, example alt index score of 64 out of 100 show underlying positivity, recent public discourse reveals significant caution primarily due to AI strategy concerns and competitive pressures. So a great job of just understanding overall sentiment over the last month of what people are talking about Apple. Is it good or bad? Uh, right. It gave us a green column and a red column. Key positive sentiments, key negative sentiments, top 5, uh, recurring themes. That's great. Uh, I'm gonna try just one more thing here. Uh, I'm gonna zoom out and I'm gonna say, I'm gonna say make, uh, let me copy this and I'm gonna say, make this more interactive and visual. All right. We'll come back to that. Let's go back and see if our Chicago-pedia got even more— got even more Chicago. Let's see. It did. Fantastic. Now we have a dedicated sidebar column for the teams, Bulls and Bears. Doubles. Yeah, I'm from Chicago. I love this. This screams out like, you know, '90s Chicago. I love it. It says tall buildings and stuff. The lake. Dibs. Dibs. You know the rules. Yeah. Throw your chair out. Reserve your parking spot on the street. This Chicagopedia, I love it. Right. And the cool thing is, if you didn't know, the code is all here, right? So, yes, you can render everything live inside Google Gemini 2.5 Pro in the Canvas feature. but if you did want to take this offline, you can copy and paste this. Sometimes it won't work, just copy and paste because you might need to install some certain libraries. Uh, sometimes it will, it depends on kind of what languages are being used. This is strictly HTML. So I think in theory I could just copy and paste this, put it on a website and it would be good to go. Uh, and y'all, should, should I publish this Chicagopedia, this Chic— Cagopedia. I don't know. This one's, this one's kind of fun. I like this one. All right. Sandra already says she's going to rewatch this episode. So let's see. Jackie is asking, great question, Jackie. Can it get past logins on social platforms? No. So all we can do with, you know, those different tools that Google was using to look at the web. That's the open web, right? So anything on social media for the most part is closed web. So even on Twitter, right, you're like, oh, everything's public. Well, you have to be logged in because, you know, certain— there's certain restrictions specifically on social media that a lot of scraping sites or, you know, tool use, kind of tool use or internet use tools from AI large language models cannot pick up that information. Great question though. Love Chicagopedia. Yeah, I do too. All right. I have so many examples, y'all, and I'm surprised that many of them are working. So let me scroll through here and I'm going to try to find maybe something that's a little more impressive. Okay, here, here's one. I think this could, I think this could be good. All right. So I'm saying, let's go ahead and launch a new window here in Gemini 2.5 Pro. All right, I got way zoomed out. So I'm saying, uh, create a visual memory game or interactive quiz that will help me learn and memorize this content. All right. Uh, so then what I'm gonna do is I'm gonna go to the Your Everyday AI page. I'm gonna click on episodes. I did mention this, but you can go read, watch, and listen to anything on our website. So, uh, you know, I'm going to our episode from Monday where we did the AI news that matters, right? If you didn't know, you can listen to the podcast on the website for free. You can watch the video for free. We have a little write-up from some of the key points, you know, and then we have a complete transcript as well. All right. So all I'm going to do, I'm going to copy and paste all of this information. All right. I'm going back into Google Gemini. I'm just pasting this and I'm saying, create a visual memory game or interactive quiz that will help me learn and memorize this content. All right, I'm going to click enter and let's see what happens. All right, so talk about business use cases, right? How about making an onboarding fun? You know, you have all these long, boring onboarding docs, right? Make a fun game out of it. Right. So this is what I'm doing. I love finding new ways to learn. I love learning with NotebookLM, the audio overviews. I love NotebookLM's new mind map feature. But I'm always finding new ways to learn. One problem with AI, it's making it harder for me to retain information. I learn way more per day than I did pre-LLMs, but I also— that means I forget more. So I'm always looking for new and better ways to learn and retain important information. So again, think you can use Gemini 2.5 Pro to automatically curate, uh, you know, certain information that you might want to that you might want to learn. In this case, I'm just using a transcript from a podcast. All right, so let's look, see what it did. It's done. Oh gosh, this is gonna be embarrassing. All right, so it created a quiz. There's 15. 15. Hey, you guys wanna do first couple questions? Livestream audience? All right, let's just do a couple questions together. See if you tuned in. Let's see if you tuned in Monday. So it says AI News Quiz. It, it just, for you all. This looks pretty good. It's got this kind of purplish background, very like Web 2.0. There's hover animations. It's pretty slick. It looks nice. It's not some ugly, janky, you know, 1990s looking quiz. It looks really good. All right. So livestream audience, let's play along. We'll just do a couple of questions. So it says the deterministic aspect mentioned in Microsoft's agent Flows aims to reduce issues like, is it high cost, hallucinations, uh, language translation errors, or slow processing speed? What do you guys think? I'm gonna take a sip. All right. I'm gonna guess AI hallucinations. Yay. It said correct. Cool. All right. So it works. That's the thing. I just one-shotted an interactive quiz based on, I don't know, couple thousand words and it took like a minute. If this doesn't change how you think you and your team can interact with even your own internal docs, I don't know what else to say. Next question. Livestream audience, who's gonna, who's gonna get it first? All right. Uh, this is meta, but not meta like Facebook. Meta as in we're using Gemini 2.5 Pro to ask about Gemini 2.5 Pro. Uh, what key feature allows Gemini 2.5 Pro to process extremely large amounts of text, audio, images, and code? Ooh. Ooh, this one's a little tricky. So cross-layer transcoder, deep reasoning agents, deterministic logic, or 1 million token context window. This one's very actually interesting because it didn't just make things up. The wrong answers, which hopefully I get this right, livestream audience, get your vote in. The wrong answers are actually key terms from other announcements from Claude and from Microsoft. But we're asking about, we're asking about Gemini 2.5 Pro. I believe it's the 1 million token context window. Oh, good. I got it right. All right, let's do one more. All right. It says OpenAI is reportedly nearing a funding round of what massive amount, potentially led by SoftBank? Okay, this one's actually a little tricky. Because there's a total amount. So is it $40 billion, $10 billion, $20 billion, or $33 billion? There's actually a total amount of funding, and then there's a certain amount of funding that SoftBank is reportedly on the line for. But that's actually two different amounts. One amount is if OpenAI does successfully transition from a nonprofit to a for-profit, and the other amount is if they don't. So there's technically three terms, a total fundraising term, SoftBank, A, if they do convert to for-profit, B, if they don't. So the question is, OpenAI is reportedly nearing a funding round of what massive amount? The amount of the funding round is $40 billion. All right, we got it right. Uh, the cool thing is I can say something like, make it more, you know, make it even more interactive and, um, detailed, maybe some slight animations, make it look and function better, right? That's the coolest thing. I didn't write a single line of code. I don't need to. I can control this with just natural language like, yo, LLM, make this better, make it shinier, make it blue, make it harder, make it easier, make it for pros, make it for amateurs, right? Create a graduated model, right? First, You know, give me 10 questions that are much easier, uh, then, you know, help me level up or, you know, turn it into more of a video game, right? There's so many things that you can do. All right. Uh, I'm gonna give this a second, uh, to finish. Uh, let's check in. Oh my gosh, look at this, y'all. So our, our Apple sentiment analysis, remember I just in natural language, what did I say? I just said make it more interactive and visual. It improved it by a lot. So there's, there's some things that didn't fully render, right? So there's some code that says like more rounding, but overall it made this look much, much better. It gave it kind of this, these, these gauges and barometers with, you know, certain, you know, filling. It just made it look much better. So these are toggles, little toggles, even though there's not a lot of information in them. So Really good. Really good. All right, let's see. All right, it's already done. Our news quiz is done. It added a status indicator. Okay, now it's actually hard. I don't know. What episode number and date was featured in the AI news summary? Oh gosh, without looking this up, what was it? I think it was 493. Oh good, I got it right. Okay. So, okay. Unfortunately, the, the, the status indicator did not light up, but I could change that. All right. So, uh, very impressive. Should we do one more, y'all? Should we do one more? Should we wrap this up? Um, let me know. You guys, you guys all got this right. I'm looking at our, at our live, uh, at our live comments. You guys got it right. You, you must have all watched this episode. All right. Uh, all right. You guys said one more. Let me just go ahead. Let me see if I can get something that I think is maybe impressive. Okay, cool. Let's do this. All right. Uh, we're gonna do one more quick one here. So this, you know, we talk about use cases. I just randomly threw one out. I'm like, how about you make your internal documents a little better, a little more fun, right? So here I'm saying, uh, essentially you're an HR expert using Gemini 2.5 Pro. Uh, you know, hey, uh, you, you, you work at IBM. Uh, create a, a, a manual with standard operating procedures, uh, for new employees. So essentially I'm saying create an onboarding form for new employees at IBM, and then also, you know, an 8-question quiz that covers key SOP elements. Ensure all recommendations are based on real IBM training methodologies. So I'm wondering if it's actually going to go pull this and find this information from the web. I guess I'll have to verify this later, right? Just because human and expert in the loop doesn't mean I need to do that live, right? I'm not going to post this and say it's perfect and working, but you'll see already one thing I love That's a little different with the canvas inside Google Gemini versus some canvas features or functionality in OpenAI or Anthropic's Artifacts is it can create multiple kind of canvases. Is it canvases or canva? Right. I think it's canvases, multiple canvases at once. So the first one is just this onboarding material. Okay. So it's creating a an SOP with preboarding, day 1, week 1, role clarity, compliance and ethics. Right. So it's doing the kind of like boring, right? All right. Here's your, here's your text-based, you know, content here. And then this probably won't be done yet, but it's— let's see. There we go. It's already done. Okay. It created an interactive version of this simple you know, 10-step SOP onboarding for new hires at IBM, right? So it has our onboarding SOP. It's interactive. It has these tabs. I can click weekly tasks, weekly schedule and tasks, and there's toggles with dropdowns. Here's week 1, foundations and setup. Week 2, role clarity and tools. This is really good. It's all interactive. It works. And here's a quiz. Is this quiz? I don't think the quiz is going to work. Let's What is the primary focus during the first week of onboarding at IBM? Leading a major product, completing essential compliance training and initial setup, presenting a strategy report to senior leadership. I'm going to guess it's the middle one. All right. So it doesn't say unless I have to like click. Okay. There is a thing to submit the quiz. So I'm just going to click one. I'm wondering if it's going to tell me what's right and wrong or give me a score. This would be very impressive. A multi-step quiz embedded inside an accordion. All right. So it didn't tell me which ones were right or wrong on this one, probably because there's no database. And then it has a checklist as well. This is very cool. So this is my onboarding milestones checklist, right? And when I check it, it says 2 of 10. I check one more, 3 of 10. Let's see what happens when I finish it out. Boom. Says 10 outta 10. Very impressive, y'all. All right, we covered a lot. I know this episode was all over the place when we talk about different use cases, uh, for Gemini 2.5 Pro. So I'll say this, it's not perfect, right? It's not perfect. Uh, the ceiling is high, the floor is finicky, right? But as long as you, the human, you the business leader out there, are paying attention, are being patient, are properly prompting Gemini 2.5. And you know, you might have to dip your finger a little bit into Google's AI Studio. Extremely, extremely powerful, state-of-the-art, multimodal, multifaceted large language model in Gemini 2.5 Pro. And the use cases are tremendous, right? It's, it's, actually baffling how many, even just what we went over here live, right? I didn't really plan these. I didn't refine them. I wanted to give you guys just, just the nitty-gritty, right? Let's see some mistakes. Let's try to improve it a little bit. But if your brain isn't churning, if one of these didn't hit home, you got to check. You got to check for a pulse, y'all, because what we just showed in this one quick episode, podcast audience, I'm sorry. I know this one was a little bit more visual. I know I didn't do a great job at, you know, describing everything, but, you know, make sure you go watch this one. But if you didn't get at least one idea on how your business, how your role, how your department can fundamentally change by using Google Gemini 2.5, you got to rewatch this because it's in there, right? So think, what public data do you have? How can you make old documents? How can you bring them to life? Right. In the same way that we talk about large language models becoming multimodal, right? I think businesses also need to start taking that same approach even for their own internal document. We don't live in like, we don't live in a text-based world anymore, right? We can create games, we can create interactive quizzes, we can create, you know, visualizations and business dashboards now with zero coding knowledge, right? Before, you might have to have a team of developers, some people in BI. Right now, you can just copy and paste. Uh, that was one of the things I wanted to do, but we ran out of time. Copy and paste a bunch of data, create a, a, a business dashboard, uh, right? And you're off to the races, right? You already have ways that you can instantly, uh, use generative AI to grow your company and your career. That's what it's all about. All right, I hope this one was helpful, y'all. Part 2, uh, again, maybe you just listened to this one for the first time. Make sure to go back one episode. Listen to part 1 where we go over more of the details, the bullet points, everything that's kind of under the hood, how the model works, all that. But hopefully in this live demo example, sometimes they work, sometimes they don't. I hope this was helpful and I hope this is sparking some ideas in your brain on how you can use not just Gemini 2.5 Pro, but just large language models in general, right? If you're not already using generative AI in large language models, day to day for every aspect of your business. You've got to rethink how you are working. You need to rethink your role, rethink your department, rethink your company, rethink what it means to be a knowledge worker. That's what all of us are. All right. So it starts here, but you need to go to youreverydayai.com, sign up for the free daily newsletter. We're going to be recapping today's post. You know, if some of y'all shared some examples, maybe I'll throw one in the newsletter as well. So thanks for tuning in. Hope to see you back tomorrow and every day for more Everyday AI. Thanks, y'all. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.