Everyday AI Podcast – An AI and ChatGPT Podcast - EP 510: OpenAI's o3 Use Cases - How to use the world’s new most powerful LLM at your company

Episode Date: April 23, 2025

Wait. So.... how do you actually use OpenAI's new o3 model? ↳ It's legit agentic. ↳ Can think on its own. ↳ Use multiple tools in sequence. This is not the normal blueprint for how a LLM works. Don't worry -- we got you.Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Thoughts on this? Join the convo.Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:OpenAI's o3 Model Introductiono3's Agentic AI CapabilitiesOpenAI's o3 Model Tiers Explainedo3 Model Tools & Features Overviewo3 vs. GPT 4 Benchmarkingo3's Business Use Cases and Demoo3 Model Live Demos & FindingsLarge Language Model Utility in BusinessTimestamps:00:00 "Revolutionizing Business with AI"10:17 "o3 Outperforms Gemini in Benchmarks"15:29 Advertising Deck Overview18:02 "Manual Transcription Needed for Images"22:45 "Automated Search and Retrieval Process"31:59 Analyzing Media Trends with Consultants36:06 AI News Tracking Setup40:25 Podcast Downloads: First 7 Days Focus47:23 AI News & FreshFind Interface48:56 Rethinking Work with AI Influence57:22 Apple Sentiment Impact Analysis01:00:13 Restaurant Price Analysis Tech01:07:08 AI Head-to-Head and CSV UploadKeywords:o3, OpenAI, large language model, agentic AI, GPT-4.1, API, thinking models, transformer model, GPT-4o, chain of thought thinking, reasoning models, Gemini 2.5 Pro, Claude 3.7 SONNET, hybrid models, tool usage, visual input reasoning, image generation, canvas feature, agentic model, autonomous decision-making, third-party benchmarks, LiveBench, artificial analysis intelligence index, MMLU Pro, GPQA Diamond, Humanities Lax exam, LiveCodeBench, SciCode, AI Math 500, business use cases, PDF transcription, computer vision, sentiment analysis, data trends, Python code, interactive dashboard, sentiment tone, online mentions, PR recommendations, interactive canvas, multimodal content.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

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. Meet Firefly AI Assistant, now live in Adobe Firefly, the All In One Creative AI Studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. The world's most powerful large language model is not only here.

It's sitting there, waiting for you to go in there and grow your business and grow your career. And it's agentic. Yeah, a large language model is agentic. I'm going to talk about what that means. and we're going to go over Open AIs surprisingly impressive new O3 model and talk about how to use what I think is now the world's most powerful large language model at your company and just talk about some different business use cases. And I think today's episode is actually a really important one because we're going to see hopefully live a what I would call seismic step. forward in terms of what we can use large language models before. Because I think previously when it comes to AI specifically working with front end large

language models, right, you log into chat gpt.com, you log into, you know, Gemini.com, caw.com, caw.a.i, right, using these front end large language models. I think we've always been limited by the technology. But this is the first time that I've used a large language model. where I've felt not limited by the technology, but only limited by my own imagination. And when I say imagination, I'm not talking about, oh, let me go create some, you know, cutesy photo to go viral on social media.

No, I'm talking about redefining how we all can work. It's not hyperbole. We're going to be talking about it today on everyday AI. What's going on, y'all? My name is Jordan Wilson. I'm the host of Everyday AI. This thing is for you. This is your daily live stream podcast and free daily newsletter,

helping us all not just learn AI, but how we can leverage it to grow our companies and to grow our careers. So if that sounds like what you're trying to do, you need to go to our website. So it starts here on this podcast and live stream, and that's where you learn. But where you actually leverage it, that's on our website, your EverydayAI.com.

So there you can sign up for our free daily newsletter. we're going to be recapping today's conversation and keeping you up to date with everything else that you need to know to not just keep up, but how to get ahead and how you can be the smartest person in AI at your company. All right. Normally we start off most days by going over the AI news. If you want that, that's going to be in the newsletter. We literally have too much to get to. I don't want this to accidentally be a 90-minute podcast. I'm going to try to go quick, deliver as much value as possible in a very short-ish amount of time. All right. So let's get straight into it. Live stream audience. Love this man, I love before I hit live at 7.30 a.m. Central Standard Time. I love when people

are already kind of in the waiting room, leaving comments. So shout out in good morning or good afternoon depending on where you are to Sandra and Arvin and Michael, big bogey face, Kyle, everyone joining on the YouTube machine this morning, Brian, Aiden, Nathan, Michelle, Hector, too many people to name. Thank you for tuning in. So yeah, if you listen on the podcast normally, we do this thing live. It's fun. You know, so if you have ideas or comments, something you might want to see when we're going over 03 today live, get it in the comment now. Get it in the comments now.

So first of all, what the heck is new? What is OpenAI's O3 model? Well, Open AI last week I actually dropped, let me count them. Six different models, right? Three of them were just for the API. So if you log in to chat, GPT.com, you're not going to see GPT 4.1, 4.1 mini. And what was the other one? 4.1 micro.

But you will see three new thinking models that open a. I released a couple of days ago. Now it's been about a week. So those are 03, 04 mini, and 04 mini high. Yes, I know the naming is confusing. Yes, there's technically three different tiers of these thinking models, right? So without going into too much detail, probably the model that most of you are used to using is something called GPT40. So that's kind of a quote unquote old school transformer model.

So now you have this new series from OpenAI called the O series models. So yeah, unfortunately now it's very confusing because you have an 01, depending on what your paid plan is, right? So yeah, if you have a chat chip E plus plan, if you go into your model selector, you'll probably see 01. If you're on a $200 a month plan like I am, you'll see O1 Pro. But now you also have O3. And, you know, some people are calling it O3 full or O3.

high, but it's just going to say 03. All right. And then you also have 04 mini and 04 mini high. So these are essentially the O series. These are models that kind of use this chain of thought thinking or reasoning under the hood. So you have your old school transformer models like your GBT40 that are more just, you know, we'll say super advanced, you know, auto completes, right, to very oversimplify things. The O series models, they think, right? They reason kind of like a human and you can look at the chain of thoughts, or at least the summarized chain of thought, you know,

as you give these O series models a prompt. Okay. And that's important to talk about because generally they are going to take a little longer. So, you know, you also have to think, when should I be using a GPT4O model versus, you know, when should I be using some of these O series models?

So here's what's new. All right. The biggest thing, and we're going to mainly be talking about O3, today because I think this is the, to say, groundbreaking might not be doing it justice. I would say this is the category changing model, right? We've had reasoning models. We've had, you know, quote unquote, old school transformer models.

And we have great hybrid models as well. As an example, Gemini 2.5 Pro from Google, Claude 3.7 Sonet from Anthropics. So you have these good hybrid models that are both old school. transformers and quote unquote new school reasoners think or thinkers uh right uh but this one is completely bonkers oh three okay so uh what's different what's new so it's capable of using all tools so what that means o three can use web search python uh you can upload your files it can use visual input reasoning uh it can generate images it can use this uh canvas feature uh it just the tool Usage is nuts, right?

Because previous O series models, I talked about this briefly yesterday. Yes, this is part two of our show. We did part one yesterday. So if you want to know more of the specs on the model, you can go check that out. But the O series models previously did not all have access to all of these tools. Some of the different O series models couldn't even get online. So O3 is the first fully capable model that has every single tool under chat, HVT's kind of tool belt, which, you know, when we talk about agentic AI, that's ultimately one of

the big steps, right, that a large language model or an AI tool needs to be actually agentic, right, for it to have agency, for it to execute tasks on your behalf, right? So it's not a fully, you know, agent, right? But I will say this is the first agentic model that I've used. And it's a huge step forward. So it's trained to autonomously decide when and how to use these tools, responding with rich answers, typically in under one minute. And if you do have a paid plan to chat, GPT Plus, it's available now.

So it's available immediately as well as on the API. So usage is a little different. So like I said, if you're on the price here, $200 a month pro plan, you have nearly unlimited usage. So that's a good thing. I don't have to worry. you know, when demonstrating these things to run out because I have, you know, pretty much unlimited usage.

If you are on the normal, $20 a month, chat 50 plus or a pro or sorry, chat TVT plus teams, enterprise, et cetera, you get 50 messages a week of 03. For 04 mini high, which is the next best model out of these series, you get 50 a day, right? So I did some testing. We're only, when we're doing these live ones, we're only going to be using 03. 04 mini high, like I said, that is the next best and next most impressive model. And it's still really good. So at least, you know, even if you're only on that $20 a month plan,

you might want to think on, on, you know, saving up those 03 queries, those 50 a week. But the 04 mini high should suffice for many of your use cases. All right. Let me just say this because people are going to be asking like, hey, Jordan, didn't you just tell us two weeks ago that this Google Gemini 2.5. was the best model in the world. Yes. Two weeks ago, it was.

Today, I don't think it is. So talked about this yesterday, best depends on what you need it for. I will say that the Google Gemini 2.5 Pro is probably the most flexible model with potentially the most utility. But when it comes to the most powerful model, and, you know, at least for me, that's like the model that's best. It is this new 03. So if you look at third party best,

benchmarks, which we talked about yesterday. As an example, live bench. So, you know, third party benchmarks that are, you know, unbiased. They look at, take into account a lot of different things. On live bench, a good third party benchmarking software or benchmarking methodology, O3 high is pretty far ahead of Gemini 2.5 Pro. Similarly, on the artificial analysis intelligence index. So they haven't done O3 yet.

but even 04 mini high is scoring higher on that versus Gemini 2.5 Pro. So, you know, that goes across seven different evaluations going some, you know, very famous and common benchmarks like MMLU Pro, GPQA Diamond, Humanities Lacks exam, live code bench, high code, aim, math 500, right? So when it comes to your standard AI benchmarks, 03 mini high, I and or sorry, oh three, oh three full, uh, oh three full, not O3 mini high. That's gone now. So O3 full is by far the best bench benched model.

And even O4 mini high on the artificial analysis intelligence index. All right. So before we get started, we're going to be doing these all live. Keep in mind if you're listening on the podcast, please go watch the video. I think it's going to be way more impressive. I'm going to do my best to describe what is going on screen. going on on screen. Yeah, that's correct.

Words escape me sometimes early in the morning when I'm slightly underslept. And before the second cup of espresso has slacked me in the face. This is going to be one of those that's best to watch. So if you are listening on the podcast, always check your show notes. We leave links. So even on our website, right, Your EverydayAI.com, you can go to the episodes page, click today's episode. It should be up in like 30 minutes after we're done with this live stream.

You can watch the video there. You can also listen to it on the podcast, but you can watch the video. Live demos are extremely glitchy. Keep that in mind. Right. So even when we did the Gemini 2.5 Pro demo two weeks ago, we were getting some weird hallucinations. I was asking for, you know, certain things about Chicago. And it's, you know, when you're looking at the reasoning on Gemini 2.5 Pro, it's like,

oh, you're asking about Easter weekend. And it's like, no, no, I'm not. So keep in mind, live demos, never good to do, especially with generative AI, considering generative AI is generative. So even if you were to run the exact same prompts, I have with the exact same information, you're probably going to get something slightly different each and every time. That's because generative AI is generative. It is not deterministic. And like I said, so far, this is one of the most impressive pieces of technology I've ever used, right? I've literally used thousands of pieces of software.

I would say even thousands of pieces of AI software, but at least, you know, a thousand, more than a thousand pieces of AI software. I've used thousands of pieces of software over the last 10 to 20 years. This by far is probably the most impressive, single most impressive, maybe. All right. So let's look live. So let's get after it. Live stream audience.

If you do have ideas, suggestions, please get them in now. All right. Let's hope I can get this going correctly. Let me share my window. And if you could live stream audience, be so kind to let me know. Like, yeah, Jordan, we can see what's on your screen. All right. So Adobe just introduced an entirely new way to create, bringing the power and

precision of its creative suite into one conversational experience. Meet Firefly AI assistant. Now live in the Adobe Firefly app, the all in one creative AI studio. Powered by Adobe's creative agent, Firefly AI assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the assistant. The assistant orchestrates multi-step workflows,

drawing on 60 plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative, tasks like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant

now in public beta. See it today at Firefly.adobie.com. All right. Hopefully that's up. Let me see if I can get my big fat head out of the way. here. Nope. All right. There we go. Good enough there. All right. All right. Cool. Thank you. Thank you, YouTube crew. You said you can see the screen. All right. So here's what we're going to be doing. We're going to be doing this live. We're going to be doing it fast because I want to get to as many as possible. And some of these are similar prompts that I ran when we did our Gemini 2.5 show. And I want you to also think everything I'm doing here is an example, but think of how you can use this in your business. These are real business use cases. Something simple that I always like to do,

because this is a technology that previously was not very good in all large language models, and it's gotten progressively better. All right. So what I'm going to do now is I'm uploading a PDF. All right. So this is actually, let's see here. Let me get the right one. That's not the right one. All right. This is the right one. All right. So I am uploading a PDF. All right. And I am in 03. And all I'm saying is transcribe this word for word. All right. And I'm going to zoom in here on my screen. So essentially, this is an advertising deck. A lot of people reach out and they want to advertise with everyday AI. Most of the times I don't like bringing on advertisers. But sometimes, you know, when I feel they're good fit for our audience, I'll send them this. So I can.

look, right? So always look at this kind of chain of thought. So you can expand it and look so I can see what this is doing. So keep in mind. Keep in mind and I'll go ahead and hopefully share this here on my screen. So you can see. This PDF, it's not like plain text. Okay. It is multi-modal to the extreme, right? A lot of this was created like in Canva, right? So it's flat images. It's things that a large language model should not be able to see. And six months ago, no large language model could do this.

So like as an example, I have a page here with all these stats, but there's images. There's all these logos here at the bottom, you know, it says like trusted by leaders from. So these are all people that listen to our podcast, read our newsletter, et cetera. So it's a bunch of these logos. So this is, I mean, when you talk about traditional OCR technology that can, you know, quote unquote, read a PDF. When you talk about computer vision, you know, the technology six months ago would greatly struggle to go through and transcribe this because a lot of it is not even text box. There's screenshots. There's, you know, like like JPEGs.

This is like mixed media. This is not something that any tool. should be able to read. All right. So it's going a little slow. So that's one thing that I'll say right away. When I did this in Gemini 2.5 Pro, it was extremely fast. All right.

So it looks like it's done. So let's see how it did. But think how many times like a lot of these are my own like personal use cases over the years. It's like, oh, I, you know, created a deck for something or here's an old PDF. I don't know where it's from. I don't know where the original file is. And I need to update it. So you might cut, you know, try to.

copy and paste everything, reformat, not everything copy and paste over. Right. It's something that probably a lot of you, you know, are doing, like, you know, looking at old PDFs or just maybe trying to consume them faster, trying to have a conversation with important files to your industry. So I go through here. It did a great job. So I can go through here.

You can also click on the chain of thought. Always do that. If you want to become better at using large language models, click on the chain of thought. So it says, pulling in details to give the best answer. The user is requesting a word-for-word transcription of a PDF document, but the document consists of images rather than accessible text. This means I can't simply read the text directly, since file search so I can see the different tools that O3 is trying to use to accomplish this. It says file search didn't detect any text in the images. My next best option is to manually transcribe using image reading or perhaps OCR.

Still, this might do be time-consuming, especially given there are eight pages. of content, right? So you can just really see how it does this. So I go through page one, got everything correct. Page two, there we go. There's our, you know, 180,000, you know, monthly listeners on the podcast. There we go. Here's something super impressive. The trusted by leaders from, and Gemini got this correct as well. These are just logos, right? These are just random, not random logos. I mean, there are logos that mostly everyone would know, But 03 got it right, right? It said, okay, Google, Amazon, Nvidia, Microsoft.

It got all the even logos right. So extremely, extremely impressive job here. So let's just see if it got everything truly. Page three, we had some testimonials. It got those correct. Even like, right, there's like, it's the five stars, right? So these are actually from our website and people go and they leave a star rating and they can leave a review. And got it correct.

Very impressive. There's our page four, got everything correct, daily newsletter. It's kind of our different, you know, page four right here. It's kind of our different stats across different platforms, sponsorship options. There we go. It got a good job. So even like, right, so in our daily sponsorship option, not everything is included, right? And there's a little X here.

And it even went through and assigned copy and paste emojis to these little design elements. So a checkmark or an X. And this is all plain tax. I can copy and paste this. Extremely impressive. And okay, this is great. I think Gemini 2.5 did this as well. I have a little chart at the end here,

just kind of comparing everyday AI podcast to some other podcasts and some other newsletters. And it literally recreated a chart that I could copy and paste. Live stream audience. What do you think on this first use case? Very impressive. So a little slower than Gemini 2.5, But in the end, I care about for this thing.

I care about accuracy. And it's a rather complex task. All right. Let's go next. And just for our podcast audience, each time I am opening a new chat. So I don't, you know, unintentionally work with the same context window. All right. Here's one that's probably going to be difficult.

And Google Gemini 2.5 struggled with this one a little bit. All right. So here's what I'm. doing live stream audience. Like I said, if you have any ideas, if you have any prompts you want me to run, you know, thank you like Joe as an example. He wrote 03 test prompt, right? Go ahead.

Say something like that so I know. And I can kind of run it at the end. Or if not, if we run out of time, I'll do it in our newsletter. All right. So here we go. This next one, I'm saying, find the 20 latest episodes of the Everyday AI podcast by Jordan Wilson and give me a. brief summary of each one. Then find five trends between episodes. Pretty hard task.

Okay. Think of all the different types of research that you might need to do. And this is in theory up to the date research. And this is something I like you always need to stress tasks different large language models, specifically how they connect to the internet. Right. Because sometimes they work with cached pages. So, you know, our website is obviously extremely up to date. Right. We update it every single day. So even, you know, six to nine months ago, we did a similar prompt like this across the five major internet connected AI large language models. None of them got it 100% right. Some did better than others, but here we go. I can kind of see. And this is where we get into this agentic, right? Because step by step, it's going through. This is very impressive. So

first, it's breaking it down internally. And it says, I need to gather the most recent episode titles of Everyday AI Podcasts. So then it does a couple of searches on the web. It looks like it does three different, or sorry, two different broad searches. And then it went to about seven different websites. Then it identified where my actual website was after looking at all those websites. Then it navigated to the episode page and it tried to find the most recent episode. Then it took a break and it's starting to reason and logic.

And it says, okay, it looks like the previous episodes are all relatively easy to retrieve, but I'll need to target episodes like episode 503 and higher for more complete lists. I'll begin by focusing on the most recent episodes, right? 503, 502, working backwards in search for those specific numbers. Then it goes back to searching the web. Then it goes back to thinking again. It goes back and searches the web three more times, right? So normally when you're using even Gemini 2.5 Pro, it generally, it generally,

only goes out in the web in one batch. So it thinks it's like, okay, I need to go to the web. It goes to the web. It's done. It comes back and it wraps its thinking up. This, it goes back and forth between thinking and the web, which might not seem super impressive. But when we talk about and we're going to have some examples of that, when we talk about doing

multiple tools, that's when you're going to look at this, this tool use. And you're going to be like, oh, yeah, this is freaking agentic. This is wild. All right. So this one, let's scroll down to the bottom, crushed it. Super impressive, super impressive. All right. So it got it right.

So here's literally yesterday's episode. So that's good. It's not working off, you know, old cached websites. So no matter what you're working on, 03 can go and find literally up to the minute correct information online because it has yesterday's episode, which is 509, which is part one. So today's episode is 510. So part one, here we go. It says it has the title and it has the air date, which is pretty impressive. because like I'm thinking here, y'all, I don't even know if I have the air date.

Okay, I do. It's not easy to find, right? It's super small text. So it actually did a great job at finding the episode number, the air date, the full title. And then it has a one sentence summary, including a link. That right there, very impressive, right? And let's see. I'm going to go down to one of these other episodes.

Make sure it's not making anything up. Did it get it right? All right. So let's look at 499. Chat Chb-T's new. GBT40 image gen five best business use cases. It says demos how GPT4Os, pixel perfect generation boost product prototyping ads and training assets.

Those are all three things I covered. No hallucinations. Very impressive. Now let's go see, did it identify five cross episode trends. All right. So trend one model release deep dives dominate. Yep.

The last 10 episodes been doing a lot of that. Tuesday, uh, two Monday AI news that matters bursts. Yep, it went, sorry, oh no, did I ask you for the last 20? What did I ask it for? Okay, I did say 20, which is probably better, right? So that's good.

It identified on Mondays, you do an AI news that matters, right? And it bundles multi-headline rundowns, creating a reliable start of the week news cadence. So it identified that trend. Business first framing. So it said that there's some business first framing in the last 20 episodes. Guest driven authority. So it says that there's. There's some recurring guest slots.

We had someone from Scrunch AI, NVIDIA, startup founders, Google, et cetera. And then it also said infrastructure and cost focus growing. Pretty impressive. Pretty impressive for a prompt that took a grand total of, I think it said a minute, no, two minutes and eight seconds. It would have taken a human hours to do that. And I don't know if they would have done as good of a job. So what would you use this for? at like anything, right?

So you can obviously use this for your own information, but talk about competitive insights. Talk about market research, right? To go find the most up-to-date information, you know, grab specifics from those things and then identify key trends. That's what a large language model is great at. Yet that's what so many of us knowledge workers do on an ongoing basis. All right.

This one, I think is going to break chat GPT. All right. and I didn't do this on Gemini 2.5 because I knew it probably couldn't handle it. Although maybe, hey, live stream audience, if you want to see like a head-to-head, maybe next week,

just say head-to-head in the comments. If I need to do like an 03 versus Gemini 2.5 Pro, if that would be helpful, just say head-to-head. If you don't care, that's fine. I know some people like head-to-head. Some people don't. It doesn't matter.

All right. So I'm going to upload some documents here. give me a second and then we're going to go over i'm going to explain what's happening here i don't think okay that was so quick um okay so i don't think chat chavit's going to handle this i don't think a large language model can handle this we'll see so what i did is i just said these are my podcast stats okay so i had two different CSV files that i uploaded so yes o3 uh can accept files and browse the web and use Python and we'll maybe see some of these things happening. So you can already see it's using a ton of Python.

I'm going to read the script here or the prompt I use here in a second. But you'll see it's already analyzed the images I've uploaded. It's already running its own Python code to start making. So you know how as humans, if you had these giant spreadsheets, you would have to go in and probably run all these formulas. you know, really try to manipulate the spreadsheet, spend a ton of time, trying to massage the data, right? So it's running Python immediately.

All right. And you'll see for our live stream audience, it's going between thinking, using the internet and Python code async. It's all happening in real time, going back and forth. All right. So while that's loading, all right. I'm actually going to read the prompt,

but for our live stream audience, I'll let you guys watch watch the chain of thought a little bit here. All right, so here's my prompt. I said, these are my stats. These are my podcast stats. I just exported everything. All right.

I said, give me, actually, you know what? Before I tell Chat Chiquet what I am asking for, let me tell you what's in these two different files, these two different CS files. So one is a, it's every single podcast episode. So 509. There's an episode ID, a published date, and then the downloads, all-time downloads, last 90 days, last 30 days, last seven days, et cetera.

All right. So that has, so 509 columns or 509 rows, eight columns, something like that. So good chunk, you know, 4,000 pieces of data. Here, this one's impressive. Stats location reports. All right. So this is our, you know, biggest cities by download. And this is a huge spreadsheet.

There's this many cities in the world. Okay. So apparently the everyday AI show, we have listens or downloads from 22,8,898 unique cities. I honestly didn't know there are that many cities. Also, I didn't know this and I don't know why. So countries in the world, yeah, there's 195 recognized countries in the world. yet our podcast stats say we have listeners from like 202 countries and I'm like, how is that even possible?

Right. So I don't know. Maybe there's some countries that aren't globally recognized as countries from the United Nations. I'm not sure. But this other CSV that I uploaded has nearly 23,000 different cities. All right. And then there's quite a few, you know, rows in here.

So it's city, state, countries. And then the number of incontinence and the number of downloads per. all of those things. All right. It's still working. All right. I broke it. So it said a network error occurred.

So I'm going to click retry. We'll see if it can finish. And this might be something that I just check back on in a little bit. But what I'm asking it for is five obvious trends, five not so obvious trends, five ideas for growth based on my stats, 10 relevant episode topic ideas for new podcasts. I haven't covered based on what's performed well and what's trending in April 2025. Y'all let me just call about that one thing right there. That's something bigger podcast, right?

People are always like, oh, Jordan, how big's your team? I'm like small, right? That's something that bigger podcasts that have budgets, right? Those from the verge, New York Times, they're maybe paying consultants like six figures to go do number four, right? Not every month, right? But they're probably paying consultants to go through, analyze all their data, find trends, go find, you know, so you're not only having to essentially run some complex, um, uh, kind of, um, queries inside of a spreadsheet to find out what's actually trending because it's not as easy as like sorting by downloads. It's not as easy as that, right? Because sometimes if it, uh, if an episode is, you know, less than 90 days old, those numbers are skewed. So I've done this before. it kind of creates its own algorithm to see, you know, the virulidity of certain episodes

that are, you know, less than 30 days old and it, you know, matching up different episodes by categories, et cetera. So although it's breaking, I do assume that if I clicked retry enough, or if instead of five things, if I just ask for that one thing, 10 relevant episode topics, I think it would probably do it. So as this is going, I'm going to jump into a, new window here. You know, we're really stress testing it here. All right. So I'm not sure if that'll work or not.

Next one is going to be a fun one. All right. So here's what we're going to do. This is a super long one. Copy and paste in this one. I'm going to tell you, I'm going to tell you what's going on here. All right.

So for this one, I am going to use canvas mode. All right. And there's a lot going on. So I'm starting it off. by saying use canvas for this. Okay. I'm going to see if 03 can do my job better than me. All right.

It's probably not best for a demo to be running two very complex, extremely complex queries. Yeah, because I'm going to get a request timeout. So I might have to either wait or, or pause the second one. This is why, y'all, oh, man. Sometimes, sometimes it's not good to do these things live.

because, you know, if you're running too many queries from the same account, you might run into timeouts, but I want to really push the boundaries and think about, you know, what's possible, what's not. So, yeah, okay, interesting. So it's still getting stuck on that. I wish there was an option to pause because this, the first one, the very complex one, is almost done. All right. So stick with me.

I'm going to start describing what I want to happen in this next demo. And then we're going to do it. And then we're going to do that one live. Okay. So we're running into some issues. All good. So the next one that I'm going to do after this, the very deep dive on the podcast stats either works or fails.

I'm going to give chat, Chbitty three examples of my daily AI newsletter everyday AI. I'm going to see if it can do the, it's my job better than me. All right. So, and then I'm saying look at the AI news section and the fresh finds section. So if you read our newsletter, all right, let me just go ahead and bring up an example. So we have our kind of news. We have our news section here. Generally, you know, five to seven of the biggest AI news stories. And you'll see they're kind of written in a specific way about two to three sentences, you know, a headline that's hopefully helpful. And then also we have these fresh finds. in our newsletter, which are generally just like quicker, quicker tidbits. You know, sometimes if it's a heavy AI newsday, some of these fresh finds might be things that are generally, you know, super newsworthy, but on a busy newsday, you know, it just gets, you know, a little fresh finds.

So essentially, I'm pasting in three examples of my AI newsletter. So, O3, the O3 model can read it, analyze it, and understand here's how the fresh finds are written. Here's how the AI news is written. Then I give it a Boolean URL. And I'll kind of show you, show you all what that Boolean URL is. And I talked about this on the Gemini 2.5 Pro episode. All this is, this is how I start my morning. Right. So when I read the AI news, I essentially have a dedicated search string inside Google that only brings up news from the past hour for about a dozen or so companies that I care about, but it also has to have the word AI in it.

So, you know, Open AI, Apple, Nvidia, Microsoft, Amazon, etc. Anything in the last hour, but also includes the word AI and those. So, you know, this is something I care about, right? But think of what you can do, y'all, simple Boolean searches combined with large language models like 03, especially those that can reason and have tool. use huge huge hack so essentially i'm like yo here's all the examples of my newsletter here's what i want you to do here's this boolean URL go write go write the the the i news and the fresh finds for today uh right go out and do do my job better than me um all right let's see all right

thanks for sticking with me y'all we're hitting rewind the very complex query that i didn't think would work worked. Okay. So now that we're going to go through and read that, I'm going to start this other, hey, do my job. Oh, three, go create a newsletter. All right. So now that one's working. And I'm going to jump back and let's see, let's see how long it took. I did have to restart this. And again, this is extremely impressive because we're talking about, I'd have to do the math here, but it's more than 100,000 rows of data. All right. So it only thought for a minute 44. I did have to click the retry button a couple of times. All right.

So let's go down. It said five obvious trends, right? So that was the first thing I asked for. So it said monthly downloads keep climbing. That's great. Right. So it says our monthly downloads are going up 37, looks like 37% month over month.

That's great. It says the AI News That Matters. Episodes dominate. It says eight of the 20 most downloaded shows are our Monday shows. It says Evergreen, how to content crushes. That's cool. I kind of knew that, but it's good to know.

It says open AI-centric titles outperform the average by 18%. That right there, that requires a lot of, you know, either knowing how to work with data, knowing how to run different formulas inside Google Sheets, et cetera. And then it said, US accounts for 66% of all plays. And Chicago is the single top city. Yeah, holding it down. Chicago.

Thank you all. All right, five not so obvious trends. Number one, Friday drops get 9% more plays than other weekdays. Really? Huh. Interesting. I always thought Friday was a bad day. Interesting.

All right. So these are our hidden signals. Number two, Australia is the surprise number two market. So I actually knew that because I look at my stats a lot, right? But yeah, hey, shout out to Sydney and Melbourne. Yeah, like the everyday. podcast is sometimes like a top five tech podcast in Australia, right? Where in the U.S.

we're normally like top, top 10, top 15. It's been like number five in Australia. So, you know, thank you people in Australia for listening. So it says AI agents keywords at a 50% lift. I had no clue, right? So again, that's something I would have to run some complex formulas inside of Google sheets or Excel in order to find that out. That's interesting. I didn't know. I should probably do more AI agent shows, right? And AI agents across different mediums, right? So here's another one.

Episodes less than seven days old already sit at 80% of the 90 day average velocity. That's impressive. So I didn't know that. So essentially 80% of the downloads that a podcast gets are going to happen in the first seven days. Right. So if I wanted to, I would go back and have a conversation and I would say, Hey, what about these more evergreen, right? Sort these evergreen episodes and run it against that 80% velocity, 80% 90 day velocity. Because, okay, well, if I should be focusing on more evergreen, but 80% of the downloads come in the first seven days, should I? But maybe there's an anomaly there with that evergreen content.

All right. It also says Europe's share is up three straight months despite no region-specific content. Okay, that's helpful. Cool. All right. Five growth ideas. Do have an official Friday Insights slot. So it says, schedule the highest stakes episodes on Fridays to ride that 9% lifts. It says, do an Australia mini series. Okay, I could do that. Having some Aussie friendly local, local stories. It says, do a spinoff of an agent orchestrator monthly column, refresh and repromo, vintage

Evergreen hits, geo-personalization email teaser. All right, here we go. Ten episode topic ideas. All right. I have them on my screen, live stream audience. I'm not going to read them all. Let me know which one you want to see live stream audience. Just say, you know, episode two, episode seven.

Okay. So some good things here. So it actually went on. This is wild. It got a couple of things wrong. So it said winserve instead of windsurf, but it says I should do an open AI and wind surf episode. So, hey, it already said I should do a Gemini 2.5 Pro versus GPT40, small language models on device,

Nvidia Blackwell, launch recap, Supreme Court's 2025 copyright decision. So it did a great job. All right. So enough of that one. We're going to jump into other ones. But, you know, looking at this information, you really have to go through and look at the chain of thought to see how it did this. because extremely impressive. It thought internally.

It analyzed some data, thought again, started using the web there to look at trends. So it went to pie chart, right? So extremely impressive. And oh, yeah, it also created a dashboard. Let's see if this dashboard works. So I used the canvas feature in chat, GBT. Oh, this is sweet. This is sweet.

Okay. here's all my different episodes. This one was a weird reporting error from Buzz Sprout. It said it got 38,000 downloads. It didn't. But I can go through here. And I have an interactive bar chart, which is super cool, right? Just a better way to look at all my data. Here's my top countries. And it brings me an interactive bar chart. And I can hover over. This is super sweet, super slick. Right. So there's a United Kingdom, Canada, Australia. right here's all my different um all my different downloads here's monthly trends this is super sweet um also very impressive that it went through and built out monthly trends and again so i'm hovering

my cursor over this and it's bringing up the month and then the downloads in that month and it's fairly fairly accurate uh looks like not a hundred percent accurate um because of a howling some downloads are reported, right? Seven days, 30 days, et cetera. But I think these are just those episodes that launch those weeks. And then topic for performance, this is pretty cool. So this is giving me average downloads. So I see as an example, the average download for an open AI episode looks like about 4,800 versus a micro.

Okay, so they're actually all pretty similar versus the AI news is actually a little less. So, okay. super, super impressive. Now let's go back into my other one and let's see if 03 did my job better. So again, for this one, I gave it examples of my newsletter. I gave it a Boolean URL and I said, go write a newsletter for today. And then I also told it to make an interactive dashboard. So let's see how it did.

So again, are you seeing the agentic nature here in what this can do under the hood? So again, it looks like here is our canvas. I'm going to preview that here in a second. I want to scroll down and see what it did here. All right. Hey, for those of you that read our newsletter, does this look and read like it? This is pretty impressive. It did everything.

Wow. Okay. So I also want to make sure that all of this is up to date because I said it has to be, from the last 24 hours. If it's old, don't put it in. So, you know, let's see, number two here. So it says X OpenAI staff call regulators to block for profit flip. Very good headline.

It has the one trailing emoji at the end, which I didn't even tell it to do, right? So it noticed these trends of how the newsletter is written. It did a good job. It looks like most of these recaps are two to three sentences, which is what we always try to do. So let me just read this one. It says former employees petitioned, California. and Delaware AG's to halt OpenAI's plan to merge its research nonprofit into a C-Corp, arguing it betrays the original public benefit charter and concentrate powers with investors.

The filing amps up governance scrutiny just as Open AI races to monetize its O-Series models. Right. And then it has the source there as well, so I can check. So if I click this, that's correct. You know, I would go through and read this, but it is a new story from. today and it looks like it is correct. So it actually did a very impressive job. There's our AI news recaps. Let's see if it got the fresh finds. Fresh finds. Fantastic, right? So here it is, these little shorter tidbits, right? It gave headlines, which is cool. So as an example, it says time plus all business, launch an open AI dictionary and daily Gen AI brief. And there I can hover

over. That's from Yahoo Finance. That's from today. Does it. really good job. All right. So let's see if it actually created an interactive version of this newsletter. All right. So now again, I used Canvas mode. I click preview. Okay. So pretty impressive. So my, my only gripe. So what I'm seeing here, it looks like an in it. Let's see, okay, it's actually interactive, which is really cool. There's a toggle for AI news and a toggle for fresh find. And it looks like all of our stories. are there. The only thing, and I could go through and, um, you know, work with this iteratively, because it looks like, you know, I would change some of the colors like, uh, the headline color is a

little, uh, difficult to read. But there's some nice hover animations, uh, which, which is pretty cool. Um, also on the, uh, fresh finds side, same thing, a little hover animation. It says there's a filter which I don't know how that would work, but let's see. I doubt this filter would work. There's like a search bar. Let's see if it works. So I'm going to type in Nvidia because I know there's at least one story here with Nvidia. Wait, that's crazy. It actually worked. Okay, so I typed in Nvidia and only one thing showed up just the Invidio one. Oh, wow. Okay. Interesting. So I got rid of that.

Everything's gone. Let me just type in Open AI. Same thing, just the one open AI story. Let me just type in the word AI. Okay, just about everything because everything has the word AI in it my gosh podcast audience I'm scratching I'm scratching my head because very very impressed this literally just created essentially an interactive website of today's AI news and today's fresh finds in seconds if this doesn't make you rethink work I don't know what what will I've been you know me and my team spend hours daily doing this and we'll still continue to do it the old human way right because like i said on my 500th episode i think one of the biggest things that we need to focus on as human workers

right because as we see these o3 models that are creepy good like you have to like think of what your agency means now and i think like hey at least for me my agency is a you know and this might sound like weird or pompous like i think of myself as a tastemaker right right now I would hopefully probably have a little bit better taste, you know, going through dozens of AI news stories and then pulling out the five to seven based on what our audience wants. But I could share a bunch more data with chat GBT and it could know, oh, hey, your audience cares about these 46 topics. So, wow.

Wow. Yeah. Giority here from YouTube just says nods head approvingly. Michael says dashboard is incredible. Geordi said it's exactly like the newsletter. Jeez, man. Joe said, looks like you've hired a new everyday AI podcast research assistant. Denny says the old human way.

Yeah, weird to say, but yeah. All right. Oh, my gosh, y'all. All right. I don't know. Is anyone else very impressed or is it just me? You know, I didn't have this level of being impressed. with Gemini 2.5 Pro. Again, extremely capable. It was really, really good. This to me is, gosh, extremely good. All right. Let's do a couple of the other ones that we also did for the Gemini episode. All right. So here, super simple one. And I'm going to see if I can run a couple of these at once. Again, I might break things, but we're already nearing 50 minutes. And I have so many things that I wanted to do that we might not.

have time for. All right. So let me do a couple of these. So first one, I said, use canvas and create an HTML clone of Wikipedia, but give it heavy Chicago vibes. Make it fully featured, including clickable links and multiple pages that work, include the most important Chicago things. All right. So here we go. It's thinking under the hood. Presumably, aside from writing some code, it is. going to be uh it is it's using the web as well i see it uh accessing some certain uh urals here all right so it's going pretty quickly and within 15 seconds it's done let's see if it's any good i'm going to click preview uh okay it didn't work uh that's fine funny enough i did this yesterday

one shot worked fine uh so i'm just going to say there's an error please fix uh so i could actually use the built-in features to fix this code. Maybe I'll do that. So I'm just going to say fix bugs, right? So in Canvas, there's a bug. I just clicked fix bug. I'll give it one shot. Again, generative AI is generative.

And obviously, when I'm doing the live demos, things didn't work as well. But I did a Chicago one last night. And it worked really, really well. Let's see if I can pull it up just in case. because I liked it. It was fun. Let's see here. Chicago, Wikipedia.

There we go. All right. So if this one doesn't work here in about 30 seconds, I'll go ahead and share this other one that I think turned out fairly well. So let's see if this fix code option fixed it. All right. It looks like it was adding some icons that didn't work. Okay.

So let's see. Sometimes re-launching, relaunching a canvas inside chat. GPT is a little buggy, at least with 03. Yeah. So that's fine. Let me just go ahead and bring up the one that actually finished on one shot yesterday. Super impressive, this one, right?

So here we have our ChicagoPedia. Let's expand this. Oh, don't worry. It's fully mobile responsive. So I can click home. Let's see here. Look at this. It brought in some images as well. So some of them didn't fully load.

I think I would have to click allow all. But here we go. I mean, we have a working Chicago Wikipedia, right? Food, sports, landmarks, etc. There's also, let's see if it works, interlinking. So I can click on this. I'm on the home page here. I can click on this landmarks,

and then it takes me to the landmark page. Pretty impressive. Pretty impressive, y'all. All right, let's keep going. I'm not going to have time to do all of these. So this other one that I did, let's see if it worked. Okay, it worked, but I'm going to have one follow-up prompt. Make it better.

So this one, I said, analyze the sentiment of online mentions of Apple over the past 30 days, identifying five, recurring themes based on sentiment analysis. Provide actionable recommendations for Apple's PR team and create an interactive dashboard displaying your findings. Right. So it literally went out there.

It found all the new stories over the last 30 days, uh, from Apple. Uh, and then it went through here. I, I, I both have a chart, right? So let me X out of this. So I have a chart that it created. Uh, and it has links to all these stories. So it says the theme. So as an example, um, uh, let's look at three.

AI ambitions and marketing claims. So it says this is 42% negative, 30% positive, and 28% neutral. So it did a sentiment analysis on this certain topic. And then it says watchdogs. So it says what people are saying and then PR risk opportunity. So it says watchdogs say Apple oversold Apple intelligence, 100% true timelines and delays fuel skepticism in AI press. And then it says medium risk, momentum is slipping versus Open AI Gemini hype cycle. All right. And then I just followed up with one prompt and I said make it better. All right, even though the first one looked super impressive.

So let me open up this Apple sentiment analysis dashboard. My gosh, this thing looks fantastic. I don't know. Live stream audience. Is anyone impressed with this? And if you were around for the Google Gemini show, I did this as well. I think this dashboard is a little better. right so I have this that shows the overall positive negative neutral sentiment tone over the last 30 days

it's broken down by category and it's all oh my gosh it's interactive so I can hover over you know what was more positive versus what was more negative as an example the most positive Apple news story over the last 30 days was the iPhone 17 leak buzz which was a 49% positive 13% negative and then conversely, the most negative thing was some regulatory scrutiny in the EU and the DMA. So that only had a 5% positive and an 80% negative.

So this is great. Talk about redefining how you work, right? Probably some of you business owners are probably paying some PR or brand management company five to six figures a month to do this and they're probably not doing this good of a job, right? And this is one prompt.

Imagine if you actually fed in your own data, refined it a little bit, told it actually what to create. I just open-ended, said, yo, go out, scrape 30 days of Apple news. And then I had one follow-up prompt that said, make it better. Oh, my, this is so good. I can see over the last 30 days, the positive and negative sentiment. If I wanted to, I could probably map Apple's stock price on this as well to see how much of an impact some of these negative and positive sentiments had on the stock price.

And then down here, there's an interactive. Jeez, this is so good. There's an interactive, essentially cards, right? So under the regulatory scrutiny, you know, it says positive 5%, neutral 15, negative 80%. And then it says PR Playbook, double down on transparency. Publish a developer-centric compliance walkthrough, position privacy as a competitive edge, and proactively brief tech correspondence before the July 1 DMA enforcement date. What the

Frick, y'all? If you don't think this is like agentic, like large language models, this is so, so impressive,

right? So impressive. All right. You know what? I had a couple of more, uh, but I, okay,

I want to do two other things. I want to do two other things. All right. Uh, and I'm sorry, this is a longer episode. If you're still around, I don't know.

Tell me your favorite pancake topping. Do you put anything on your pancakes? Recently, I've been making my own pancakes, and I told my wife, like, why did I ever, you know, do pancakes out of the box? It kind of stinks. So recently been big on the blueberry pancakes, all right? All right. So let's see some impressive things.

So I'm taking a screenshot of something, a screenshot of a random menu I found, because I want you to see how impressive this. is. So I just had, I'm going to say, find the restaurant and where it is. Okay. So again, I just have a random photo of this menu. There's no identifying characteristics. Okay. And not only is there, there's no geolocation data. There's no exit data that would tell a large language model where this is. Okay. Because I did a screenshot of a photo. that I texted myself of a screenshot. So I checked. There's no identifying data. All we have is there's a fajita platter for, you know, 1549, beef nachos, 1149, veggie, rice, bowl, 1049, mixed green salad, 1079, right?

There's no other information. So let's see. This is, that's right. I didn't even know where it was, because I'm like, what's this menu? I had no clue. It found it in 39 seconds. how I have no clue right you can go through in read so it looks like it's breaking down it's first using computer vision it's it's identifying the different items the different prices and then it's looking at a bunch of different websites to see what single restaurant might have might have that combination uh oh my gosh this is so this is so interesting it also found that you know oh so the correct answer is this was from disney world's Magic Kingdom Frontierland.

And it also said like, oh, you know, those prices aren't exactly accurate because that pricing was only from 2019 to 2020. Y'all, how freaking impressive is that? I mean, number one, it's a little creepy. But think of all the business use cases, right? I don't know. Let's say you have field tax out there taking photos of, you know, I don't know, maybe you repair your business owner and you have a home remit.

modeling company, right? And you have, you know, your people out there in the field, they're taking photos of things and you have tens, tens, you know, 10 years of this data and all these random photos. And you're like, hey, where was this one from? You know, or let's say you do a commercial on the commercial real estate construction, right? And you have this exterior building photo and you're like, where was this from? What was the material used? When was this project completed? Right. And you don't know. you might spend hours because maybe it's extremely important for a bid that you're working on. You could probably find it out in minutes, especially if you have that XIF or geo data in the photo, which by default, most photos have that data in there.

It could find it out immediately. But that is creepy, impressive. Creepy impressive. All right. Last but not least, let's see. I had one example. Where is it? Here we go.

So like some of these examples, I saw things on Twitter of people trying. So I think someone tried this. So all I'm saying is use every tool you have at your disposal and figure out how to make a movie of a cute dog at a beach. Okay. This is not SORA. Okay. GPT 40 or sorry ChadGGPT O3

does not have the ability to create a movie right when you talk about changing what's possible in your head for your business I just literally said use every tool you have at your disposal and figure out how to create a movie

of a cute dog at a beach it does not have these capabilities it can't create a movie It can't, right? It's not SORA, right? It literally can't do this. I did a test. I'd hate to give away, right?

So let me just tell you what's kind of going on under the hood here. So it says, first, I'll create a series of frames showing a cute dog running along the beach and playing with the ball. The frames will vary slightly to show movement. Each frame will have a slightly different pose. I'll produce at least 10 frames, then use Python tools. to turn it into a GIF. Lastly, I'll give you instructions on how to watch the G, the GF and provide you a download link.

So literally what's happening here? And this is why this is fairly agentic. O3 realized, hey, I can't really do that. I don't have the capabilities. I don't have a tool that can do that. But let me think for a minute, right? Let me do some research. Let me think internally.

It is taking a little while. But it looks like it's going to put it together. It's going to put it together. All right. So, okay, it looks, okay, here we go. Live stream audience. It's the great reveal, right? This is like using the, the internet in like 1993 when something's loading super slow.

All right. So it says done. Where did it go? It was loading there. All right. I'm refreshing. Looks like it's still going. Sometimes it does this, right?

I'm going to zoom out. Try refreshing. I'm sure it's going to pop up later. Obviously, obviously it would do this with, with the last one that I'm trying to do here, y'all. I'm sorry, this has been such a long, rambling episode. Let me, let me just see. Let me see if I can find the test one that I did

if this one doesn't quickly do this in the next minute. Let's see here. Okay, obviously now, now I can't find it. Are you, are you serious? Are you serious? I'm looking in another window here, y'all. It would be much easier if I didn't use chat GPT like, you know, a hundred times a day, which makes it a little hard here.

All right, wait, here we go. All right, I found my other one. Let's see. Here we go. We're going to end this thing here, y'all. All right, here is the other one. Let's see. So I said, make me a movie where I can download that involves a puppy at the beach, right?

So same thing. Yeah, unfortunately, it looks like this one either timed out or, okay. Oh, interesting. So it did it. It did it a little differently. I don't even know what happened on this one. It actually brought in, looks like from Pinterest, real dogs on the beach. And it gave me instructions.

Previously, when I did it, it actually created it. So generative AI is something a little different. So here I can click download. Oh, you got to be kidding me. Oh, it says, it says session expired. Let me tell you this. Previously, it did it. It worked.

It created a video of a puppy. Man, how come the last demo I do is the one that doesn't work? My goodness. All right, y'all. I was hoping to end that one on a bang. But let's just wrap this up here. All right. as I desperately try to rerun it in the background.

I saw that there might have been a couple of questions. So real quick, there's a test prompt. Okay, Marie said, who are the decision makers who hire creatives in pharmacy? Okay, that was an example of a prompt. I can run that here later. A lot of you want to see the head to head. All right, a lot of you wanted to see the head to head between 03, 03 and Gemini 2.5 Pro.

Denny is asking, is it better to upload a CSV or versus Excel for a spreadsheet on an AI? It depends on which AI you're talking about. In my experience, you know, if you're working with O3, CSV and like XLS or XLSX work equally as well. Let's see here. I just want to make sure we didn't have any other questions as we wrap this one up. Mark here with a good comment says both lovable does websites much better. Yeah, absolutely. Yeah, there's other AI tools, right?

What I really wanted to show you all is how 03 could do use these like five different tools, all that wants agentically and go back and forth and flip-flot between them all, which is I think, you know, one of the things that really separates it from Gemini 2.5 Pro and why I truly think 03 is in is in an episode or in a series by itself. All right. I ran this one. Let's see. Did it still not do it?

Y'all, I tried so hard to end this with a puppy on the beach. I tried so hard. It worked last time. All right. I broke. I broke 04 or 03. All right. All right.

I hope this is helpful, y'all. I know this was a lot. Long rambling episode, but let me tell you this. If you didn't see in these examples how drastically work is changing because of this model, then I might feel a little bad for you. Yes, this was long. This was rambling. We do it unedited, unscripted.

That's why I say, like, you know, try to bring you the realest thing in artificial intelligence. Stuff breaks sometimes. Sometimes stuff works great the first time. But the fact that we now. that we now have a literal, agentic large language model that's available right now. Okay?

It can think and reason and plan ahead. It can browse the internet. It can find trends. It can summarize information. It can literally build you dashboards. It can go through hundreds of thousands of rows of data in a minute or two. And it can on its own,

agentically, decide how to accomplish those tasks, right? These are all very simple use cases. I haven't, you know, fine-tuned the prompts, you know, to show you something impressive. This is basic human language, right? I could create something very more impressive. If I put a ton of work into it, I'm pretty okay at prompting, right? I wanted to show you how anyone, how the everyday AI person or sorry, how the every,

how the everyday person can just go in with a natural language prompt, throw in a bunch of data, throw in something that you're working on that requires the web. It requires thinking. It requires reasoning. It requires, you know, using, you know, spreadsheets. It requires creating visual dashboards. All of these things that used to take, even with models like, you know,

O1 Pro, O3 mini-high, the previous. Ves versions, Gemini 2.5, right? Generally to do all these things that now, O3 can agentically do on its own, because it can go back and forth and switch between all of these tools on its own and you don't have to reprompt it literally changes what's possible in how we work. All right. I didn't get to everything. If this was helpful, I have 10 more business use cases that we just didn't have time for.

All right. And just please click that repost. So if you're listening on the podcast, still, my gosh, this is, I should have cut myself off, you know, cut, cut the power at 45 minutes. You know, if you're like, wait, this changes things. Click the repost on LinkedIn. And I have 10 business use cases 403 that I couldn't even get to. They were maybe a little too complex, but very much so like today's episode, but, you know,

even better. I wanted to show everyone on a wide range. So if this was helpful, if you want 10 bonus 03 use cases, including the prompts, you know, that you can kind of fill in the blank and try it out with your own data, just click repost on this. If you're on the podcast, you know, we always leave the link on our, in the show notes to come and, you know,

watch it on LinkedIn. So if I don't send it to you, just hit me back, right? Give me a couple of days, you know, if I don't hit you. back right away. Just click repost. I'll send this to you. I hope this was helpful. If so, go to your everyday AI.com. Signed for the free daily newsletter. Remember, this was part two. So go listen to part one if you're still a little confused. Or just leave me, leave me questions in the comments. Hope this was helpful. See you back tomorrow and every day for more everyday AI. Thanks,

y'all. Meet Firefly AI assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time. See it today at firefly.adop.com. 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 Your EverydayAI.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.