Ep 883: How To Make Slides with AI: Hands on Comparison With the 9 Most Popular Options (Replay)

No matter your role, no matter your industry, no matter your experience level, chances are you still spend a lot of time doing this one thing: making slides, creating those decks, right? So whether that's toiling in PowerPoint, copying and clicking in Canva, or working on a template in Google Slides, just about everyone who sits in front of a computer screen spends time Creating slides, and this was one of the areas of the early generative AI phase where the tools really kind of struggled mightily for years, and there was a lot of hoopla over how there wasn't a lot of good AI options. Then, kind of silently yet kind of quickly, we got a host of solutions. But I think most people overlooked just how good many of these AI tools are at generating slides because we essentially saw an avalanche of model capabilities all hit us. At once. And I think that reasoning models in agentic systems took away the spotlight as slide generation kind of arrived at the same time. So no one really took notice. Well, we did. And that's why we're going to put AI to work on Wednesday and take a look at at least 9, well, maybe 10 of the best ways to generate slides with AI, all aimed at beginners. And we're doing it all live, errors, hiccups, And all. So which is best? Will PowerPoint shine? Uh, what about the Nano Banana integration in NotebookLM? Or maybe the old kid on the block with Gamma, or the new token-hungry Claude design could take the crown. I don't know. We're gonna see. Let's find out and get into it. So here is the big picture. The capability gap right now makes it hard to keep up because I can almost guarantee, even if you, uh, watch this show every day, there's gonna be one of these tools that I talk about and you're going to be like, wait, I didn't know you could create slides that way. And it's just now by default, a lot of these large language models create slides, right? You know, it's something that kind of snuck in in terms of capabilities, you know, around 6 to 9 months ago. Like I said, about the same time that we got all of these agentic systems and reasoning models, right? But I remember telling people, Uh, like 9 months ago, people didn't know that you could just create a spreadsheet in ChatGPT or, you know, create a PowerPoint in Claude, and you can't, right? And finally, now with Google Gemini catching up to the pack, all of the big players by default can create artifacts like that if you just use the right model and ask it to. So when you combine though your dynamic data inputs and then the personalization and the customization that comes by default out of large language models and schedule those things, you can have on-demand personalized decks, right? And then when you can throw in your brand options as well, a lot of people overlook just how capable this is, right? Because I think slide generation has now turned from a task that so many people spend so much time on To now you can get an on-demand, pretty polished draft, at least in minutes. So on today's show, here's what we're going to cover. We're going to go over the pros and the cons of the top 9 options and see how they make slides and look at their output. So everything from ChatGPT and Claude, Gemini, Gamma, PowerPoint, and more. You're going to learn how the tech actually works, at least on the surface level, and go over best practices. And you'll instantly, by the end of this, if you put this into practice, you're instantly Going to be able to implement going from zero to draft by the end of the show. No more overwhelm, no more confusion. Let's get started. Welcome to Everyday AI. My name's Jordan Wilson, and well, we do this every day. It's your daily unedited, unscripted livestream podcast and free daily newsletter helping everyday business leaders like you and me make sense of the AI noise. These updates are coming nonstop. I tell you what matters, what doesn't, how to use it. You take that information and all of a sudden you're the smartest person in AI and you can go grow your company and career. Starts here, unedited, unscripted livestream podcast. But to take it to the next level, make sure you go to our website, youreverydayai.com. In today's newsletter, we're gonna be not only recapping the highlights from today's show in case you miss anything, but also all the other AI news you need to know to get ahead. And yeah, this is gonna be one of those episodes that's a little more visual. So you might wanna, for our podcast audience, Yeah, check out today's newsletter, uh, and, you know, go watch, uh, this video on our website. So we link to it in the newsletter. It's always on our website too. So let's get straight into it. Uh, and let me say this, this is meant to be a quick demo. All right. I'm not going to show you my best practices using one system. Uh, I think that, well, I hope at least that this is going to be more appealing. And, you know, this is actually, I put this out in the newsletter. You all voted for this as the thing that you wanted to learn most. I want to start doing more practical guides, right? And instead of going deeper on one system, say, hey, here's how you do data analysis in 3 to 5 tools. Here's how you do, you know, create slides or decks. Here's how you can create, you know, quick disposable apps or websites. So gonna take these a little bit more topic-based. But again, I'm doing this very generally. My prompt is gonna be overly simple. I'm not gonna be, you know, using, you know, codecs and or using skills and plugins. I'm doing something that's so simple. Your grandma could go and do the same thing. Or if you are a grandma, you can do it along with me, right? I'm making it that simple, right? The simplest prompt, uh, you know, basic paid plans, nothing crazy. So we're going to be seeing how a model can correctly follow the input because we're not just going to be judging the quality. All right. Because models by default are multimodal. You have to be able to understand How they work, the pros and the cons. So, you know, the prompt that I'm choosing is intentionally vague so I can see how these models, number one, handle it. Number two, because we're going to be indirectly testing prompt adherence, tool calling, and also taste. So again, this is not a final workflow. And if you are doing something like this, right, pick the one that you want, double down, but let's get in. Let's start live. Oh my gosh. Like, not gonna lie, this one I'm a little nervous on. I'm also at Microsoft Build right now, so I just have my little laptop. Normally I have a lot of screens, so might be a little jumping around. Hopefully not. Let's get started. Let's get started. Livestream audience, let me know, can you see my screen? Hopefully you can. All right, we're gonna go in order, and here is the tricky part. There's gonna be a little bit, 'cause not all of these are just gonna be, you know, I click, you know, put my prompt in, click go. Some of them force you through some, some different prompts and, which can be a good or a bad thing. So I'm gonna try to walk you through which ones are accepting one-shot generations versus which ones are asking me follow-up questions, right? Which I think asking follow-up questions in most cases is usually a good thing. So keep that in mind. The sample is not going to be consistent all the way across because by default, some of these systems do kind of, you know, force you Or prompt you to go through a certain series of things to get more information. All right, let's start first. So the first one we're gonna go over is Claude Design, and then we're gonna go over Claude. All right? And just FYI, these are very token hungry. So if you're on a $20 a month plan, even if you try to do these 2 comparisons, you might run out of tokens for your 5-hour limit. This also is gonna take a while, these ones. That's why I'm doing them first. All right, so Claude Design, you know, I don't use it a lot in this account. I have another account where I use it a little bit more. But all you're going to do is you're going to go to Claude.ai, and then on the left-hand side, there is a little Design tab. So Claude Design is technically its own kind of little app. It's not the normal Claude interface. Also, it only exists on the web version. There is no Claude Design if you use the desktop. All right, and then there are a couple different things that you can do. And this is a research preview. So there's a prototype, a slide deck, a template, or other. So right now, I'm just gonna go Everyday AI News That Matters. We're gonna go to slide deck. Click create. It's gonna take me through a little kind of prompt where normally I could go into a design system, add a screenshot, attach a code base, drag a file in, all that. So I'm not gonna do any of that. All I'm gonna do is put my prompt in. And I am going to be using Claude Opus 4.8. All right. All I'm saying, and this is the same prompt I'm going to be using in each instance, and this is why we're going to be carefully following prompt adherence, tool calling, and taste. Those are kind of the 3 things we're looking at. So the same prompt I'm putting in for all of them, please look up the latest AI News That Matters episode from the Everyday AI podcast by Jordan Wilson. Please create a slide deck Giving each of the main stories, uh, a one-slide recap, one intro slide, and one outro slide. So I believe in Mondays, so from 2 days ago, so on Mondays we go over the AI news that matters. Usually we do like 7 to 8 stories. I think this Monday was 7. So in theory, there should be 9 slides, an intro slide. Each news story should have one slide and then an outro slide. All right. So I'm going to send this. Inside Claude Design. And I do believe I'm going to have to wait here on the page a little bit because I do think it's going to ask me some follow-up questions because I didn't give it anything else. But I'm going to give it a second. So one thing I'll just, you know, as I go along, if I do have to wait on a certain system, I'm going to try to not bounce back and forth too much. So I might give you a little commentary as we go along. Oh my gosh. Right. Terrible. So Claude Design is great. One of the downsides, and you can do things like prototyping, right? This obviously isn't just for DEX. You can build, you know, entire, you know, UI/UX wireframes. It's a really powerful tool. It is extremely slow. And like I said, it is token hungry. So, you know, depending on the complexity, even if you're on a $20 a month paid plan, you might not even be able to kind of create the one thing that you want. All right. So now it kind of went through the tool calling. All right. It said it searched the page, fetched the page, found the information. It told me the main stories. It looks like I only had 6. That's interesting. I'll have to double check that. So it said before I designed a few quick questions so the deck looks the way you want it. So the thing that I really like about Claude Design is you just click on it. So in the right-hand side, there's options. So it's asking me the overall visual style. I'm just gonna pick whatever on these to go through them quickly. I'm gonna say clean editorial color direction, dark background with a blue accent. Copy tone on the slides. I'm gonna say keep some of Jordan's takes and opinions. The good thing is, obviously, which I love this about Claude Design, all of these kind of questions it's asking me, they're obviously custom. Right? It's not some, some boilerplate questions it always asks you no matter what. It does a really good job thinking through and asking you good questions. But I, the thing I like is you just click, you don't have to type, which is good. Uh, it says how to handle the stories. Uh, so I'm gonna say, uh, you know, 6 story slides and then branding. Sure. I'll say, uh, reference the Everyday AI, uh, News That Matters as the source. Sure. Okay. Uh, and then I'm gonna click continue and hopefully that's all. Uh, I might have to check back. I'll probably do that here in a second. All right, so next we are going to just Claude now. All right, so same thing, I'm using Opus 4.8 Max, uh, I am using, uh, thinking mode, and I don't think I'm gonna have to do any follow-ups here. So in this instance, this is just the model going out and doing its job, and hopefully it's going to correctly find this information and create a slide. All right, so we'll check back on that one in a couple minutes. Next we are going to Gamma. So Gamma is actually one of the, you know, the old kids on the block, so to speak. So this kind of when Gamma came out, it was almost in a league of its own, right? This is before Claude and ChatGPT and Gemini and even Microsoft Copilot could generate slides with AI. Gamma was one of the originals. So in my Gamma account, again, I have paid accounts for all of these. Maybe this is why the podcast should be making more money because I shouldn't be spending a ridiculous amount of money every single, you know, month testing these tools. But I do this so you don't have to, and I can just tell you honestly what's good and what's garbage. All right. So anyways, when you go into Gamma, I'm clicking Generate with AI. I'm choosing the presentation. It looks like I can't— I'm going to have to choose 8 cards. All right. Hopefully that'll work. So I do have to tell the cards by default. I can click Classic or Studio. Okay. Actually, oh, apparently I'm not on the right plan for Studio. That's fine. So that's it. So I'm just, I have again my same prompt. I'm not gonna read it each time, but please look up the latest AI News That Matters episode from the Everyday AI podcast by Jordan Wilson. Please create a slide deck giving each one of the main stories a 1-slide recap, 1 intro slide, and 1 outro slide. Go. So on this one, the thing first, Gamma is going to generate an outline. All right, I'm going to scroll through it. Looks like I can see it already made some errors. Interesting. That's fine. I'm not going to correct those. There's options for the type of text contents, the visuals. I'm just going to choose these. There are— these are preselected for me, so I don't have to do anything. You know, I can just click generate and then we're done. All right. And then I can also select a different image model. So there's an auto select model. You can use, you know, Flux, you can use GPT Image, all these. I'm just gonna click the default one and we're gonna click generate. All right, and then it is off and running, off to the races. All right, here we go. Next, we are using Google Gemini and finally, finally we got essentially artifact creation in Google Gemini. So there was kind of a way to do this prior. It was almost a little workaround. I covered this like a year ago. But really up until a month ago, Google Gemini did not have the ability to create slides. It didn't have the ability to create spreadsheets, right? It could create a link to Google Sheets. It could create a link, but you couldn't actually create downloadable assets, but you can now. All right. So I am again on my paid account. And the reason I'm saying paid account, because some of these features and functionalities, if you're using a free account, are not going to work, but if you're on the base Gemini paid plan, this will work. I'm going to— the model I'm using is 3.5 Thinking. Same thing. I'm going to click go. I don't think I have to clear anything here. All right. I'm just going to give it a second just to make sure. I'm trying to remember because I think actually Gemini, you might. All right. And I'm going to quickly check in on my other ones, make sure they're cooking. All right. We got things cooking in the other ones, not to give it away for our podcast audience. Okay, we're gonna give— Gemini is taking a sweet second. Nothing's loading. There we go. Okay, so I thought I might have had to click approve a plan like you have to do with Gemini Deep Research. It looks like it's going. I might have to check back in on it in a second. That's okay. So we actually have 3 separate 3 separate options, uh, from Google. Okay, looks okay. Apparently it's, it's ready already. So that was pretty fast by Google Gemini, FYI. And also there's actually way more ways we could have done this, right? We could have used and maybe would have had even better results using like Gemini Canvas. Uh, we could have used the new Codex Sites, right? So, you know, even from a coding perspective, which there's actually really good skills, uh, out there that can create beautiful slides, you know, with different programming languages. But again, I wanted to keep this as simple as possible, just simple input/output without having to know a lot. All right. So next we are going to Google Slides. All right. So on the right-hand side, you can just click Create a Presentation. Let me see the exact one that we would choose here. Let's see, File, New Presentation. I want to make sure I do this correctly. If it just comes up. Okay, here we go. So you— it's going to give you options. So you're going to want to click Slides. And then on the right-hand side, you can choose your different model. We're just going to go with the default here. Same thing, put the prompt in. I do think Google Slides is going to ask me a series of questions first. So before we leave and let Google Slides do its thing, it is reasoning over my simple prompt and it's creating a plan to go about the best way to do this. All right, so we'll give it a second here. Like I said, 3 different approaches that we're taking in Google: Google Gemini, Google Slides, and we're going to click NotebookLM as well. All right, so I scroll down here, it says podcast recap details. Okay, it is asking me some questions, and I like it's easy, uh, simple bullet points. So it's asking me what is the desired information density. I'm just going to say balanced. Who is the target audience? I'm going to say just business leaders, right? It's you guys. What is the core narrative or goal of this deck? I'm going to say explain industry shifting news. And then how would you like to provide episode content? Okay, this is interesting because I told it that it should be looking it up. So I'm going to say other, and then I'm just going to paste the prompt in again. Please look up the latest AI News That Matters episodes. All right, so let's see if that one works. All right, we're gonna give it a second. I'm gonna keep cooking. I'll go back and check on Google Slides here in a minute. I think that's all we need to do to get it to work. All right, next we are going to Google's 3rd entry, NotebookLM. So for this, you're gonna create a new notebook and add a source. So unfortunately, I can't do the same kind of— this is the only one where it's not gonna be able to take the prompt as is. So instead, what I had to do on the left-hand panel, if you're using BookLM, there's essentially 3 different panels. You have your left-hand panel that has all your sources, your middle panel where you can kind of chat with your sources, and then your right-hand studio panel, right-hand studio panel. So essentially all I did is I manually added that episode on the left-hand side. So the Monday episode, our AI News That Matters, we do the AI News That Matters on Monday. And then in this slide deck, I'm gonna hover over, there's a little arrow. So I'm gonna click that. And then there's a describe the slide deck you would want to create. All right, so I'm just gonna take away the first thing there saying please look it up since I already gave it the URL. So technically we're giving NotebookLM a little hint, but it's just, it just works a little bit different because it's grounded by default. All right, so I'm gonna click generate from there. Make sure it gets started and NotebookLM is off to the races. And this is going to obviously use Nano Banana to create the slides, which is pretty cool. All right, next we have ChatGPT. This one should be pretty straightforward. I'm not even going to use the best model, so I'm just going to go to use thinking heavy. So I'm using GPT-5.5 thinking, the heavy version. Same thing, put the prompt in, click go. I use this a lot, so I know ChatGPT is not going to ask me any follow-up questions. All right, next I'm using Microsoft Copilot Online. All right, so I'm on a Mac, so I don't have the full Microsoft 365, uh, you know, Copilot version, uh, that I would like to be able to demo. So I'm doing both the online version of Microsoft Copilot. I pay $20 a month for Copilot Pro, but remember, this is a personal subscription, so it works so much different than the Microsoft 365. So keep that in mind. But a lot of the underlying technology is the same. And then I'm also going to be demoing PowerPoint, but the desktop version, but with ChatGPT. Because as weird as it sounds, the PowerPoint version, even with Copilot on a Mac, the paid one, you can't do this. It, it, it can't query the web, which is frustrating. So I can't do that test, but we're going to connect ChatGPT when we get there. Anyways, uh, now we're doing Microsoft Copilot Online. All right, so, and this is the personal version. Um, I'm just using the default model, default settings, and I do think it is going to have me click to approve if I remember. Uh, so it's first pulling up all the information. Um, let's see, I'm actually gonna count this. So 1, 2, 3, 4, 5, 6, 7. Yeah, that's what I thought. So I forgot which one. One of the ones said that there was, oh no, the end says rapid-fire AI industry updates. Technically not one of the big stories, but let's count one. So the stories, let's just go over these now because we're going to have to fact-check them later anyways. So it was the number one, the Microsoft Copilot super app leak, which was announced. NVIDIA Neutron 3 Ultra. Yep. The NVIDIA AI laptops that were announced at Computex. That's number 3. Number four was the CEO stories, the CEO layoffs blaming AI. That's number four. Number five was the CEO Dario Amodei and Sam Altman talking about job risks with AI. So that's one, two, three, four, five. Number six was the OpenAI robotics launch. Number seven is the Opus 48 update. So there are seven. All right. So yeah, there should be seven. I think someone got that wrong. There should be seven AI updates. So nine slides. So anyways, I'm gonna click start. Uh, on Microsoft Copilot. All right. Also, let me know, livestream audience, as we're going along, who do you think's gonna win? I wanna know. I'm gonna take, I'm gonna take a sip here and you let me know who's gonna win. All right. Because I care about, is it factual? Did it pull everything correctly? And taste, right? Taste is subjective, but if it doesn't get the 9 slides. Right? If it doesn't properly list the 7 stories, then it technically didn't adhere to the prompt, right? And something in the tool calling did not work. All right. So now we're going over to Canva. Same thing. We're going into templates. We're going to create a new presentation. And then there is essentially, when you do that, there's a little text prompt box up here. A lot of people miss this. All right. Um, but you're just going to type it in, right? See, now it already went away. All right. Uh, you're gonna type, type in your, your prompt here. I should probably get the entire thing. This interface is a little clunky. I don't know why it's not pulling in. Okay, there we go. All right, so I've typed in my prompt and then there is a little generate button that has little stars. All right, so, uh, it's going through. It says finalizing already, even though I just put it in one second ago. I don't know if I'm gonna have to go back and check on this one. I forgot. Uh, but anyways, that's it for all of our web-based versions. All right, so now I'm going to have to, uh, remove that for a second. And now I'm going to bring up PowerPoint Desktop. So now I'm going to be sharing that. So stick with me for a second. Remember, if you want to actually see the video versions of this, Make sure to go to our website at youreverydayai.com. All right, let's go ahead. I gotta switch over, get this window. Boom, there we go. Okay, so hopefully you all should be able to see this, but now we are in PowerPoint. And like I said, unfortunately, the Copilot version of this— because there is a Copilot version, uh, it is not going to pull, uh, The information. I'll actually just go ahead and do this as I check, just so you can see, right? As I quickly go through and check the other models, make sure everything in the browser is good and cooking. All right. Everything is, everything is okay. I take it back. I did on Google Slides, FYI, I did have to click approve. All right. I'm just going through and checking on everything, making everything Making sure everything's going as we give PowerPoint here a minute. All right, not sure what the heck is happening with Canva. Okay. All right, let's go. Let's see. Okay, so you'll see here, all it did is it just created the— okay, so it says want— okay, this is so weird. So earlier when I did this with Copilot, It didn't work. Now it says I can turn this into an actual PowerPoint file with design and visuals. This is interesting because last time it did not properly browse the web. It said I can't browse the web. So this is, I, I, I, I don't know what's going on here. Uh, so anyways, it's asking me if I want to turn it into, uh, and it says, just tell me thumbs up. So I don't know. I'm going to say thumbs up, buddy. Uh, anyways. Let's see if I can start another one. All right, I can, but I'm not gonna be able to share my screen. So let me unshare this one. All right. And then let me share. Yeah, normally I can just share my whole screen, but again, I'm on a laptop, so sorry. Uh, we're having to bounce around a little bit. So now let me go back to my other presentation, which is hard to see. All right. So now, oh, there we go. I got the wrong one. Got the wrong one. All right. Y'all, I told you this wouldn't come without its, its fair share of hiccups. I'm going to have to resize this window so I can tell the difference between them. All right. This is the problem. Trying to do too many, too many things at once. On a small screen. All right, let's bring up— there we go. All right, now we have PowerPoint but with ChatGPT. So PowerPoint has a new ChatGPT add-in. So you're going to want to look in your banner bar for add-ins, and then you're going to search for ChatGPT. You are going to authenticate with your ChatGPT credentials. The cool thing here is, well, even free accounts for now work with this. You don't even need a paid ChatGPT account, which is really cool. So you just need to go through, uh, authenticate with your credentials, and then essentially on the right-hand side of PowerPoint you have a ChatGPT bar. All right, so I'm going to go to— should I go to heavy? Yeah, let's go heavy. We'll give it some time here. Let me get my prompt there, paste it in. I'm trying to remember if I have to wait here or a prove anything. Uh, I'm not sure. All right. But what we'll probably do, I'm gonna have to try to multitask here on the world's smallest screen. So, uh, livestream audience, don't laugh at me if you see me squinting like I'm 90, because I'm gonna have to be keeping, keeping, uh, keeping track of a couple different things. Okay. I think I should be fine. And the cool thing is that I really like about the ChatGPT, uh, plugin inside PowerPoint. It's really good because it doesn't just create visuals. I might just do a dedicated episode just on this because I think it's that good. We'll see what the output looks like, but let's just go ahead, give it time to cook. All right, so we're going to go back now and check in on all of our browser versions, and I'm going to give you guys my opinion. Did it follow the prompt? Did it properly bring in the information? And then the taste, right? What does it actually look like? Is it good? All right. So let me go ahead. We're going to start in the beginning with Claude Design. All right. So we have a slide-by-slide view here, and I will say this. It is both really, really good and it makes me nauseous. And maybe this is because I am terminally online and I pay way too much attention to AI, but essentially all of Claude Design's, uh, designs, all of Claude Design's designs look the exact same by default, right? So it's literally the same thing. It's always a Uh, super bold headline. There's always one italicized, um, uh, word. That italicized word is always in a light color if it's on a dark background. Literally, no matter what you do, I have no clue why, the design is the exact same in Claude Design by default. All right. Uh, it's gonna give you different color schemes, slightly different fonts, but it always follows the exact same thing. So The actual look of this, if you've never seen it before, you're going to look at it and be like, wow, this is great, right? And it is, it is good. But I am absolutely tired of seeing this design because now everyone is just, and has been since Claw Design came out, they're flooding this thing everywhere, right? A lot of people are using this for presentations and I'm like, poke my eyes out. It looks good, but I am like more so than even Delve, you know, in, in it's not A, it's B, right? When you talk about AI-isms, I think right now, you know, and there's always the purple gradients of the early vibe coding, the, the, the absolutely worst eye-gougingly bad AI-ism right now is the Claude design default. Anthropic team, please update this because you have a really good product here. All right. Anyways, now that I got that out of the way, overall, if you've never seen this before, you're gonna look at these visuals and be like, they're amazing. All right. The COVID says hot AI summer is here. All right. And then it has a nice little recap. It only got 6 of the 7 stories. All right. So it missed one. Which one did it miss? It missed— we have— okay. It missed the story of NVIDIA CEO calling out, uh, NVIDIA CEO Jensen Huang calling out the other CEOs. So, uh, for using AI as a scapegoat. So it completely missed one of the stories. Not good. All right. Visuals though, visuals look great. It has in, it has fantastic content inside, right? So if it's, it says, you know, one app, let's actually go to a better slide. Okay. I like this one. This is about Nimotron 3 Ultra, NVIDIA's new open source model. Really good. It broke out some of the key, some of the key facts and stats. It laid them out. So even though this slides, there's not a lot of variety. There's no images, obviously, because this is not an image model. But in terms of, you know, for it being no images, it's actually really good, right? Good content. I'm looking at it. Everything factually checks out, but it missed one of the biggest things. It didn't get all the stories. So bummer. All right, moving on to Claude. Oh gosh, Claude, Claude, Claude, Claude, why are you so bad at this? Uh, interestingly enough, all right, I see what happened here. Claude made another mistake. So in this instance, I know what it did. It used a skill even though I did not tell it to use a skill. And actually it made a big jump on that skill. This is why, uh, sometimes it's much better to, for a model, to force you to call a skill because it thought that I wanted to use a design skill that is technically not anywhere close to what I wanted it. Uh, so let me just scroll through, see where it did. Yeah. Okay. Yeah. So it thought I wanted to use one of the predetermined design skills, which I did not. All right. Um, So although it followed, right, if you watch a lot of our live streams, this, uh, you know, I've been, uh, testing out different, um, kind of skills, uh, across different platforms, seeing how closely they can use, you know, Python and all these other things to replicate some of our templates. Um, and it actually did a pretty decent job at replicating the template, but I didn't want it to. All right, so, um, anyways, let's see if it got all of these stories correct. So we got story 1, story 2, story 3, story 4, story 5, story 6, story 7. All right. And then it also did do a little rapid fire of what's new and what's next. That one, I'll give it credit, right? Because I said all the main stories. So essentially I always do these, the big stories, and then at the end I do kind of like a rapid fire. So if a model, if one of the outputs includes that as one slide, I think I'll let it pass, but missing one of the main stories altogether, not good. So overall, you know, hit or miss here with the actual Claude output because it did get all the stories. It added the extra roundup. I don't really care about that, but it used a skill when I specifically did not want it. And technically, by my language, it should not have. This is one of those things with Claude Opus 4.8. I don't think it's right. I don't think it's really good. I think if I would have done this in 4.6, we actually would have gotten better results. I don't think it would have used the skill. Next, moving on to Gamma. Wow, what do we got going on here? All right, got a lot of crazy visuals, a lot of neon, right? How come every time you ask for anything AI, they always use the same color scheme? All right, anyways, pretty, pretty good output here from Gamma. It does look clean. All right. It didn't go in order. Okay. Uh, for whatever reason, it put, uh, Opus 4.8 as the first story. Obviously in the order of the transcript, it's the last story. I didn't tell the model what order to put it in. All right. Uh, so it put it in its own order. All right. So we have the Anthropic Opus story there. The evolution of ChatGPT agents. Nope. That's a— I'm gonna guess a hallucination. Yep. Yeah, we just got a bunch of hallucinations here. So it looks like Gamma is not very good at properly pulling information. So yeah, it has a story about Microsoft Copilot Studio. Some of the supporting bullet points are kind of correct, but the main headline is not correct. Yeah, Google Notebook LM advancements. It looks like it's pulling things from multiple Uh, stories, or it's taking things that were bullet points, not the main stories, um, and making it a main story. And that's when I talk about things like prompt adherence is super important, right? It needs to be able to take my language, which is clear enough, right? I said the main story, so it should be able to take a look at that transcript, uh, and understand what the difference is between a main story. And I even say that in the transcript. I say our main news stories. And then at the end, I usually break it down and say, hey, here's our bullet point roundup. So these models, they have to be able to reason, they have to be able to call the right tools, and they have to be able to adhere to my prompts. So Gamma, the design is okay, not my personal taste, but it does create images. All right. So this is one of the first ones that creates images by itself because there are different image models baked in. So versus just being, you know, overly text like our first 2, you know, not bad. So I'm going through. Yeah, very, you know, super neon vibes, some hallucinations, not really following, following the prompt. So not that great. All right. Next, we have Google Gemini. All right. It rendered it as an inline, which is fine because I can download it and export it to slides either way. All right. So let's see. We have a title slide here. It does create images as well. So it did also put Anthropic Opus 4.8 first. Again, I didn't tell it what order, so as long as it has them all, that's all I care about. The next one, NVIDIA and Microsoft partnership. That's correct. All right. The next one, the Copilot super app leaked. That's fine. It uses some very random screenshots here, but it's kind of relevant. It has 2 What look like iPhones, which is funny enough since we're talking about a Copilot super app. All right. Anyways, it's okay. The slides look fine, right? It's using different infographics. It's using little icons, multiple shadings. Okay. OpenAI's pivot to robotics. And that's it. So it missed 2 stories. So didn't follow everything correctly in terms of what I gave it. So not the best from Gemini. The design was okay, right? Simple. It did create some photos in some instances, but again, failed at the basics of prompt adherence. All right, next. All right, we got another failure from Google, unfortunately. The design on this one looks better. So now we're in Google Slides. The design, I think, is actually pretty clean. It's one of the better designs we've seen so far. Uh, because it does balance, uh, kind of headings, uh, subheads, uh, bullet points, typographies. It pulled out some key, uh, some, some key points here. Uh, but it looks like the information is all just made up or generic. So I'm wondering if something just didn't run correctly. All right. So yeah, it does say it's a recap. It's a placeholder. That's interesting. Okay. So when I went through and approved it, It looks like it did not properly pull the information. But I did click approve that, so I don't know if that was my fault. I wasn't looking as closely. But regardless, when I told it to go out and fetch all this information, it just came back and made it generic. So I don't know how to score that one. Fail-ish, fail-ish. All right. Anyways, going back into NotebookLM. NotebookLM is next. And the thing is, this uses Nano Banana. So again, I gave it the URL. So I'm going to go ahead and open this on the right-hand side. So the thing with if you've used the deck creation feature inside NotebookLM with Nano Banana, it is hit or miss, right? This one I say visually is a miss. All right. At least for me personally, not loving the design. A lot of times I'll just Do 5 of them and usually 2 of them look legit amazing and 2 are okay and then one's hot garbage. So in this case, this one is, looks hot garbage-esque. All right. But we have a title slide. All right. Let's see how we did in the second one. This one, I have no clue what this is. All right. Took a ton of information. I mean, we have stuff about NVIDIA workforce, so it's kind of throwing a lot of things all together. It's almost like another recap slide. Um, here's Mike. Microsoft is collapsing its scattered AI ecosystem into a singular super app. Okay. So it looks like maybe we just had 2 overview slides. All right. That's fine. Let's see how we do in the rest of them. Uh, NVIDIA and RTX. Pretty good. All right. I'm looking at the specifics here. This is correct. All right. It's bringing in the right size of the models, pulling everything from the transcript. Uh, did a pretty good job there. Next, the AI layoff narrative. Correct. We have there Jensen Huang calling out the CEOs. It kind of made some little stats, some little charts. So, okay, good. Next, Anthropic Opus 4.8. Pretty good, right? It has this little diagnostic slider, the enterprise trade-off, the core update. Okay. So even though personally not to my taste, It actually, even though I was a little confused, it looks like it did kind of a title slide and an overview slide, which I guess is okay as long as we don't miss any of the stories. And then last but not least, this is OpenAI's software simulations. Oh, I zoomed in. Are the training ground for physical autonomy? Yes, this is the story about OpenAI moving to robotics. Pretty good. And then this last one, I'm not sure what this is. This looks like a general slide. So overall, fairly strong. NotebookLM in terms of prompt adherence didn't do the best. It is obviously only grounded in the information that we gave it. So I didn't see anything made up, anything super off topic, off topic, but it seemed to generalize some things. So not the greatest. All right, next, uh, let's go to ChatGPT. All right. So ChatGPT, we have a cover image here. It looks like the same. Uh, the same layout throughout, kind of text on the left-hand side, and then it generates an image on the right-hand side, uh, presumably using GPT Image 2. The cool thing is this also has like speaker notes at the bottom, which is pretty cool. Um, so overall I'm liking this one. And so far this might technically be the best, even though it's not visually the best, because it might hit all our points here as I'm looking. So number 1, we have the title slide. We have everything in order. The first one here is about the Microsoft Super App. Nice and clean. A great, actually a great visual on the right-hand side, even though it's not to my taste, because I just don't like all the neon stuff. It's actually a good visual. It represents, it has all these, you know, these little neon lights of all these different types of apps going into one. So it actually did a good job visualizing it. All right. Number 2, Nvidia. What's actually cool here is it used kind of the Nvidia grain in it. You know, has something that looks like a supercomputer. So the visuals are actually on point here. All right. Yep. The new PC here with the RTX Spark. Good job. The AI layoffs, really good. Has an empty boardroom with a big question mark, you know, a projector screen. Okay. This is good. I'm actually— this is the first time so far I've been impressed. ChatGPT did good. Okay. OpenAI's next frontier, physical work. This might, again, I don't like the whole robot thing, but this one looks good. This is a robot in the physical world holding up blueprints and there's a building being erected in the back. It's actually real good, even though it's not my taste. I think overall, if I showed these to everyone, people are probably voting for this one so far. All right, here's the Opus 4.8, the honesty take, and that's it. And then we have our closing slide there. So I think we got, I think we got it all. I think we had, I think we got, wait, did we miss one? Hot AI summer. We might've missed one. I'd have to go back and look unless it create a, it combined. Which one did it miss? Okay. It combined. That's, that's what happened. Okay. Because, uh, I was wondering, I'm like, I remember seeing all the information. So this one kind of Failed because it didn't give each one its own dedicated slide. So it combined the Jensen Huang calling out CEOs with Sam Altman and Dario Mati's shift on AI jobs. So overall, this one's the best, but it didn't— so technically none of them have followed directions to the T, and all of them have had so far had some degree of failure. But overall, I do think ChatGPT is in the lead for now. All right, next going to Microsoft Copilot and I'm actually surprised. This looks pretty good. This one is actually, there's some visuals here, uh, which is cool. The thing I love about Microsoft Copilot is by default you can actually edit it inline, which is really cool. You don't have to download the PowerPoint version, open PowerPoint and go edit it. So this one, pretty slick, pretty good so far. All right, so first we have the Microsoft Super App. Good. Next we have NVIDIA Nimotron. So no visuals, no photos in here. But overall, uh, branding looks pretty good. Um, correct information. That's good. Then we have the, uh, the NVIDIA and Microsoft partnership. Then we have, uh, what's this? The big picture. All right. This just looks like a recap slide, maybe. Okay. Then we have the model race. Okay. So this is about Claude Opus 4.8. We have OpenAI's robotics. Good. Another one on Claude Opus 4.8. So interestingly enough, it looks like it combined Opus 4.8. It did make a slide, just I did talk about other models on Monday and it made a slide comparing Opus 4.8 to those other models, even though that was not technically its own story. But we do have the dedicated Opus 4.8. We have the, uh, the Dario Amati and Sam Altman shift. It pulled out a Quote. All right, interestingly enough. And then a lightning round. It did round up everything. So again, not 100% prompt adherence, um, but we did cover every single story at least. The visuals inside Microsoft Copilot actually really good, even though there's no photos to it. That's okay, you know, some interactive pieces. And I love the fact that you can edit inside the app. So at least for me so far You know, maybe Microsoft Copilot's ahead. You know, ChatGPT is right there. Claude Design is there as well. So, all right, let's go to Canva. Canva, no, looks bad. It got like 3 things. I knew Canva wasn't gonna do well, but I know people would be like, you should try Canva. All right, it gave us 4 slides of information. It gave me, yeah, it hallucinated it all. It said story 1, OpenAI's GPT-4o launch, right? That's not it. So it just hallucinated a bunch of information, nothing useful. All right. Now let's go look at our 2 PowerPoint options and wrap it up. All right. As I thought, yeah, the actual PowerPoint version did not work. All right. It said it was going to, it didn't. All right. But let's look at the ChatGPT. PowerPoint version. I know this has been a long one, y'all, but hopefully it's helpful. Hopefully me spending, you know, 49 minutes on this is gonna, you know, keep you from spending 30 hours trying to figure out what's best and what's not. All right, let's go ahead and get our ChatGPT version, right? Yeah, there we go. Okay, so now we have— this is PowerPoint, but the ChatGPT Uh, add-in. Okay, so pretty good. All right, so similarly to how Copilot works, obviously because this is in PowerPoint, I can start editing immediately, uh, which is nice. The good thing is, which I'm not going to get into, this is the only one— well, NotebookLM, you have some of these, uh, capabilities, but, uh, you can edit it iteratively as well, which is really nice. Not all of them do that. And there's different modes in this ChatGPT PowerPoint integration. So cover slide, good. It also gave a nice little table of contents. My thing here, these slides are clean. These slides are very clean. Okay, probably in terms of proper spacing, alignment, so far, it's actually Claude Design is really good. So I'd say Claude Design and the ChatGPT PowerPoint versions in terms of design, spacing, typography, overall taste, probably the best. Microsoft Copilot next after that. So we have Microsoft Super App first, NVIDIA's Nimotron again using some of their colors, the NVIDIA and Microsoft partnership, really good. The AI layoff narrative with Jensen Huang. Yeah, yeah, it's doing good. It's always creating, you know, multiple pullout boxes. So it could have used, because you can use skills inside of ChatGPT. So I'm pretty sure I could have used the image to skill, but I didn't call it, right? And I didn't call it anywhere else. But I do, I do know that ChatGPT Inside of PowerPoint, could have done those things. All right, then we have— geez, just overall, I just— the design of this one is just so clean. All right, uh, so we have the open, uh, the Frontier Labs. This is the Anthropic, uh, 4H story, looks like. Then we have the OpenAI robotics story. Uh, oh no, that one was just a transition slide, so it added an extra one. That's fine. Uh, it added a feature comparison, so it made more than we asked for. And then it kind of rounded up everything with the rapid-fire recap at the end, and then kind of a, uh, an actionable call to action. So, uh, it did add some extra slides, uh, but I think it did cover everything. So that's a wrap, y'all. This one was long. There's no winner. Tell me if you're watching the video version, which one do you think? So obviously with some iteration, you could probably find a winner for you. I will say that there are certain ones that do really well. And just from doing this all the time, I obviously gave you a very simple prompt, right? And we were testing multiple things. And obviously all of these versions would have been better with some back and forth with me investing time. But that wasn't the point. I will say both from my personal experience of using all of these many times and from, you know, just my take on these, Uh, I do think Claude Design really good, uh, if especially once you start to iterate on it. Downside, by default it's always going to spit out the, the most templated thing ever, uh, even though that template looks good if no one's seen it before. But it is great when you iterate, but it's token hungry. So, you know, you're not going to be able to use it a lot unless you're on a $100 or $200 max plan. So I, I will say the tier 1 options are Uh, number one, uh, actually, let me go back and look at the ChatGPT version real quick. I kind of forgot what that one looked like. That was actually pr— okay, that one was pretty good. I forgot about that one because there's photos in there as well. Okay, so I'm gonna say that the, the top tier here is, is Claude Design, uh, the 2 versions of ChatGPT, and then the Microsoft Copilot. So out of the 10, uh, that we tested, I will say those are the first tier. So maybe that's where you start and iterate with those. Although I will say this, if you do work with one of these systems, one of these systems with enough work will usually be able to give you great outputs, right? You have to edit them a little differently. You have to go through the process. Some accept skills, some don't, right? Some accept connectors and apps, which is great. Some you can run automations, right? I could have turned this into an 8-hour course. But luckily I'm going to shut up enough in time to make it at least less than an hour. All right. I hope this one was helpful. Do you want more of these? Right? Yeah. They're a little dirtier. They're a little longer, more hiccups, more blabbing, but I like to— like the problem with AI, y'all, and other people that try to cover this stuff is sometimes it's overly polished. And I'm like, that's not how it works, right? You're going to get errors. Things take longer like this podcast. Things take longer. So if you like these longer, in-depth looks, but starting with a skill and taking that skill across different, uh, you know, AI systems, let me know. But let me know by a comment if you're listening on Spotify, if you like this, if you hate it, let me know. Uh, but please, if you haven't already, go to youreverydayai.com, sign up for the free daily newsletter. We're going to recap the too long, didn't watch version. So thank you for tuning in. We'll see you back tomorrow and every day for more Everyday AI. Thanks, y'all.