EP 375: ChatGPT’s Canvas Mode Overview: 5 things you need to know

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. OpenAI just released its new Canvas mode a few days ago, and it's undoubtedly impressive. So Canvas is a new mode inside of Chat GPT that changes and I think improves, like how you and Chat GPT can write and code together. But I think everyone's caught up on comparing it to a similar collaborative feature in Anthropics, Claude called Claude Artifacts. And in doing so, I think everyone's missing the bigger picture. With Canvas, it's actually a significant step forward in how humans and AI will work together. So I'm going to show you that and tell you the five most important things that you need to know about this new Canvas feature inside of ChatGPT. What's going on, y'? All? My name is Jordan Wilson and welcome to Everyday AI. Before we get started, I have to give a shout out to our partners from Microsoft. So the Worklab podcast from Microsoft is made for leaders who want to understand the future of work. It offers expert insights in everything from how to approach digital transformation to what it takes to thrive in the AI area. That's W O R K L, A B, no spaces available wherever you get your podcasts. All right, thanks to our partners at Microsoft and welcome to Everyday AI. Like I said, my name is Jordan and this Everyday AI thing, it's for you. This is a daily livestream podcast and free daily newsletter helping us all learn and leverage generative AI to grow our companies and to grow our careers. And if I'm being honest, I think Chad GPT's canvas mode is a great place to do that. I have been very impressed and like I said, we're going to get started here in a minute and tell you everything that you need to know about it. But before we do, let's start as we do every single day, by going over what's happening in the AI news. So Google is reportedly going to be enhancing Gemini's AI image generator with some new editing features. So the new Android beta version of the app allows users to make small tweaks to images generated by Gemini, improving customization. In a demonstration by Android Authority, users could easily change elements of an image, such as swapping a regular hat for a birthday hat on a generated image of a dog. I'm pretty sure that was something I did like six months ago. So while this feature is still in beta in the new version of Google Gemini, the early feedback suggests that the edits may not be as precise or as reliable as you thought. So this update positions Google Gemini as a formidable alternative to other popular AI image generators like chat, GPT's, dolly and mid Journey, which are constantly evolving with new tools and updates. Also right now this has just been kind of leaked out in the Android version, so nothing on the desktop yet, but presumably that will be coming to desktop. And with Apple Intelligence's image playground set to launch soon, you know, expect more out of Google and Google Gemini. I actually really love their new Imagine 3 model, but that hasn't fully made its way to Gemini yet. Our next piece of AI news Foxconn and Nvidia are teaming up to build Taiwan's largest supercomputer. Talk about a super duo there. So the new supercomputer is named the Hon Hai Kaohsiung Supercomputing Center. I probably got that wrong, I apologize. And it will utilize Nvidia's Blackwell GPU architecture and feature 64 racks and 4,608 tensor core GPUs. That's a lot of compute. My gosh. So with an anticipated performance of over 90 exaflops, this supercomputer is expected to become the fastest in Taiwan, significantly enhancing AI computing capabilities. So construction is already underway, with the first phase expected to be operational by mid-2025 and full deployment targeted for 2026. The integration of Nvidia technologies such as Nvidia's Omniverse and Isaac robotics platforms will further enhance manufacturing processes and AI applications. If you didn't know, Foxconn is the world's largest electronics manufacturer, known for providing a wide range of products to top technology companies. All right, our last piece of AI news for the day. Jeffrey Hinton and John Hopfield have been awarded the 2024 Nobel Prize in Physics for pioneering work in artificial neural networks. So the Royal Swedish Academy of Sciences has awarded the 2024 Nobel Prize in Physics to Geoffrey Hinton and John Hopfield. So Geoffrey Hinton and Hopfield are jointly honored for their foundational discoveries that enabled machine learning with a with artificial neural networks dating back to the late 70s and early 80s. Hinton is often referred to as the godfather of deep learning and developed the back propagation algorithm, a crucial method that allows neural networks to learn from errors, significantly enhancing AI model training. Hopfield is recognized for creating the Hop field network, which mimics human memory processes and illustrates how biological and physical principles can be applied to code computational systems. The Nobel Prize includes approximately $1 million cash prize, which will be shared between the. The two. And yeah, this is pretty big for these two people in AI to be winning the Nobel Prize. All right, let's get straight into it and Talk about Chat GPT's new Canvas mode. And we're going to do an overview, talk about how it compares to Claude Artifacts. But more than anything, I'm going to tell you the five things that you really need to know because I think people are overlooking what this actually means. I think people have really just been caught up on comparing it to a feature inside of Anthropics Claude called Artifacts. And I think everyone's missing the point. All right, so let's get into it. Hey, thanks everyone for joining. So, Jackie and Jason, sabbatical life. Stephen, Rolando, Tim, Josh. I'm not AI, although I do have an AI clone from hour one. All right, but if you guys have question about this new Canvas mode from Chat gbt, please let me know. And you, you guys wanted this. I asked in my newsletter, said, hey, what is tomorrow's hot Take Tuesday show going to be? You all said you wanted to hear this. So here we go. Here is Canvas. All right, so this was just released a couple of days ago and I'm going to read just a couple excerpts from OpenAI's website. I'm going to give give you all the kind of marketing speak and then I'm going to cut through it and tell you how it actually works. And, you know, we'll probably actually do something live here too. So here's Canvas. OpenAI says we're introducing Canvas, a new interface for working with Chat GPT on writing and coding projects that go beyond simple chat. Canvas opens in a separate window, allowing you and ChatGPT to collaborate on a project. This early beta introduces a new way of working together, not just through conversation, but by creating and refining ideas side by side. Canvas was built with GPT4O and can manually be selected in the model picker while in beta. Starting today, we're rolling out Canvas. So this is last week. Starting today, we're rolling out canvas to ChatGPT plus and Teams User Global teams users globally, enterprise and ed users will get access next week. We also plan to make Canvas available to all ChatGPT free users when it's out of beta. All right, and here is, I think, something that people are kind of overlooking. So a little bit more on the page. They said we trained GPT4O to collaborate as a creative partner. The model knows when to open a canvas, make targeted edits, and fully rewrite. It also understands broader context to provide precise feedback and suggestions. So that should tell you a pretty decent amount there. Just the marketing speak from OpenAI, but one thing to pull out there, this is not just GPT4O with a little feature on top. It has been trained seemingly from the ground up. Yes, it is based on OpenAI's kind of newest model, their newest general purpose model, GPT4. Oh, that's important to make a distinction here because a lot of people are wondering like, oh, what about the, you know, the newest model, the, the reasoning model, strawberry, called 01 Preview, 01 Mini. Well, that's not really what Canvas is for and I think people are kind of missing that point. I think that we're not going to see, right, all of these new ChatGPT features. I do not expect most of them to ever work with the new Zero1 model. It's because it's a different class, right? This is not a GPT model, the O1 model. It is the OpenAI01 model. So so many of these tools that we talk about, right, browse with Bing, Advanced Data Analysis and now Canvas are not going to be available and can't be used in the O1 model. So keep that in mind. GPT4O is still by all benchmarks, the most capable model in the world, right, aside from the O1 model, but it is the most capable general purpose model in the world. All right, so the biggest thing here that OpenAI and everyone is really pushing is better coding in content creation. But I think it's so much more than that. It's not just about you, the human creating something, right? We've had Chat GPT and other large language models now for almost two years, right, since the Chat GPT moment of November 2022 and then everything else kind of shortly followed thereafter. So I don't think this is necessarily even about better coding and content creation. I think the biggest thing of what people are missing here is this is a huge step, in my opinion, toward this concept of augmented intelligence, right? We talk about, you know, human intelligence, artificial intelligence, and then that thing that happens in between augmented intelligence. And I think by default, to get augmented intelligence, it's been a little tricky with traditional large language models because of the interface and how they work. Essentially it's more of a back and forth, it's not truly collaborative, right? If you prompt something inside of ChatGPT as an example, maybe you need an outline for a marketing case study that you're working on and you need an outline and you're uploading a transcript and you need an outline, right? And it got pretty much everything right. And there's 10 points in the outline. But number two is wrong. Number seven needs updated, right? So generally what you would have to do is you would have to explain that to Chat GPT or any other large language model. And what would generally happen is it would cause either a lot of back and forth or you maybe copy that information, you paste it into word processor, right? Microsoft Word, Google Docs, something like that. You got to change all the formatting and then you, from there you're like, okay, well, I'm just going to manually change this. Some of these things are wrong. So in the end, you know, people are always like, well, was it even faster, right? So this new inline editing, I think, is the biggest thing inside of Chat GPT's canvas. And we're going to show you examples of actually what that means, and we're going to do this live. But let's start with some facts, all right? I don't know. Hey, question here. I don't know when Chat GPT will get the Nobel Prize. We'll see. All right, so let's go over the five things you need to know. So talked about one of these things already. But number one, right now it is available to all Chat GPT plus users and maybe free users will get access after beta, right? You know, take, take all that with a grain of Salt. You know, OpenAI's rollouts can be very slow, right? We've all seen that. So if you're a free user of Chat GPT, I honestly would not expect this anytime soon. So if you want to use it, you might as well just get on the paid plan. The second thing that you need to know, Canvas is its own mode. Okay? You can't be. Like I said, you can't use Canvas inside when you're using Chat gbt. Oh, one model, you can't start a conversation in a different mode and then switch it over and say, oh, I want to use Canvas. If I'm being honest, I haven't seen a downside yet. I don't see a reason not to just move all of your chats or to start all of your new chats in the new GPT4O with canvas mode. Like I said, unless you are using and you want to use the new reasoning model, the strawberry model, OH1 Preview 01 Mini, in which case you would start a chat there. But otherwise I do believe that you would just always start your chats with Canvas, whether you think that you're going to need them. Or not versus using, right, the legacy model GPT4O GPT4, or using even the newest model GPT4O. To me, there's no reason to use the GPT4O mode when you should probably just be using the GPT4O with Canvas for everything. All right, here's, here's the other part that you have to keep in mind. It works with other tools. All right? And this is where I think ultimately ChatGPT has separated itself from Anthropic Claude, from Google, Gemini, and even from its, you know, kind of brother technology in Microsoft Copilot, at least Copilot for the web is because it can use all of these other features and it does it very well. So browse with. Bingo. The availability to browse the Internet and direct it to browse the Internet, right? Because when you're using Chat GPT, you need to make sure all the information that you're using is up to date and not from 2023 or 2022 or even earlier. So being able to still use browse with Bing, Advanced Data Analysis, Dolly, et cetera, being able to use all of these inside of the new Canvas mode is pretty big. And do not forget about that number four. The fourth thing you need to know, this is not a carbon copy of Anthropic Claude's artifacts. All right, let me just put the facts out now. Everyone's like, online, oh, OpenAI's copying anthropic Claude. Not really. You want receipts, y'? All. The, the new artifacts feature from Claude, it's fantastic, right? And essentially it is a side by side editor where you can talk with Chat GPT or in the fact, or in the case of Artifacts, you can talk with the model on the left hand side and then on the right hand side, in addition to carrying the conversation on the left hand side of the chat window, it's going to execute an action on the right side, right? So in the case of Anthropic Claude, the biggest differentiator there is, it can render code and that's huge. And Canvas cannot render code, right? It can write code, it can collaborate, it can debug code, it can do all those things, but it cannot render it, right? Whereas in Artifacts from Claude, you could as an example, write an entire website, right? You could drop in a screenshot of a website, you could say, write this in, you know, HTML CSS and JavaScript and then render it, and then the website will actually be there like you just built an entire website. So that is what I think is a main differentiator. When people are making this comparison. And let's also call a spade a spade here, y'. All. ChatGPT is not copying anthropic Claude with this feature because receipts. Who was technically first with this split interface, right? With interacting with a model on the left side and carrying on a conversation, and then on the right side, the model does something for you. Oh, that was OpenAI with their GPT builder back from November. Why is everyone forgetting about that? Right? Everyone's blindly saying, oh, open AI, they're just a copycat. They just copy and pasted, you know, everything from anthropic. It's two different tools, number one and number two. OpenAI was first to this kind of left, right? And having the AI build something for you while you talk to it on the left side, that's been there in the GPT builder since November 2022. It's. It's so funny. People, you know, go wild on the Internet and, you know, make all these wild accusations, and it's like, y'. All, like, this wasn't a secret. This has happened, right? The GPT builder has been there for like a year. All right? So this is not some carbon copy. All right? And then number five, which I think is the biggest one, the inline editor. The inline editor is what I think puts ChatGPT in a league of its own. This is what I think, at least for the big, large language models. This is, I think, the biggest first step toward this concept of augmented intelligence. I know, I hate the concept. It's a cheesy word, but it's the truth, right? Because every other large language model, essentially, it's just you and the model screaming back and forth at each other, even in Claude. Artifacts. I was using artifacts last night. Don't get me wrong, I love it. But it kept making these small little mistakes, and then you have to do essentially a bunch of prompt engineering. So much extra typing, a lot of extra copying and pasting because it feels like, oh, this, this AI is so smart. But it's getting this one thing wrong, right? Because in certain aspects of the word collaboration, you're not actually collaborating, you're just directing, right? And then the model spitting things out where this in line editor and being able to literally work alongside in real time with CHAT GPT, I think in the canvas mode is the biggest step toward augmented intelligence. All right, let's quickly go over. Let's. Hey, hey, Philip. Let me know. Philip said CHAT GPT can't do this. You might have been responding to someone else, but let me know what you're, what you're talking about there. All right, so here is the mode selector. So for our podcast audience, very simple. When you go into your Chat GPT account, again, assuming you are on a plus account, you're going to see a new model selection screen. So like I said, for the most part, I don't see a reason to ever not start in 4.0with canvas versus just 4.0. So make sure you always start with the right model when you get going. All right, so that's kind of a combination of number one and number two, it's available right now to all ChatGPT plus users and it is its own mode number three. So I, I, I have an example here. All right, so I'm using now chat GPT4O with Canvas. I have a screenshot here and I said, please use browse with Bing to find the 10 most populous cities in the US in 2024. All right, and you'll see here, this is an example of using the other tools. And that's important mainly because at least if you're comparing, trying to compare apples and banana chips here with Claude artifacts, Claude can't really use any other tools aside from artifacts, right? It can't access the web. There's no IM. There's a lot of essentially tools that ChatGPT has that claw doesn't. All right, so you'll see in this case, even in the Canvas mode, it went and used Browse with Bing, it pulled a list of the 10 most populous cities. All right, so before we get going, there's a lot more got to shout out Worklab. So the Worklab podcast from Microsoft is made for leaders who want to understand how work is changing. Effective leaders adapt. They stay on top of trends. They embrace any edge they can get. Effective leaders also know that the key to understanding artificial intelligence is to get better at understanding human intelligence. So for real world lessons and actionable insights to help you stay ahead, check out the Worklab podcasts. That's Worklab. No spaces available wherever you get your podcasts. All right, thanks a lot to our partners from Microsoft. Some great episodes that have already dropped. I've really enjoyed them. All right, let's get back to it. Number four. So it is not a carbon copy. Canvas is not a carbon copy. I already talked about this, but I wanted to show an example of something you can do inside of cloud artifacts because it, it has obviously its own benefits that Canvas Mode does not have. So in this example, I took this same information. Right? We're just doing a Simple, a simple explanation here where we're taking these 10 cities, right? The 10 most populous cities in the US and I also think, by the way, if you ask Claude, you get something different because it's using all the information. All right? And then I essentially say, use Claude Artifacts, please create an interactive graph with this data that you can click around on, right? Nothing crazy, but essentially what Claude Artifacts can do is it can do a lot of business intelligence. It can create interactive dashboards, it can create, you know, it can render code, right? You can run Python code, you can create websites, you can do so many things by rendering code. That is the main benefit of Artifacts. There is no collaborative, true collaborative feature. There is no ability to add it in line. That's why I think artifacts in Canvas are two completely different things. I, if I'm being honest, I only put this in the title so I could set the record straight. These are two very different things. So we should probably stop after you listen to this, we should probably stop comparing them just because they share a left to right interface and you can kind of direct the model to go do something while still continuing to chat with it separately on the left hand side. That does not make these two things the same. All right, in any ways, Claude, Artifacts created a nice little interactive graph right, where I can scroll. It gave me the top 10 cities and I can hover over and it does this little animation and then it pops out, right? I hovered over Chicago and then it, you know, gave me Chicago's actual, you know, population. So that's essentially what artifacts does. That is not Canvas. All right, so let me give an example of Canvas and then we're going to jump in live. And again, if you have questions, if you have questions, let me know. All right, I think, Philip, I finally got your question. Sabbatical life. I'll get yours too, don't worry. All right, so here's an example of using Canvas and how I think this is completely different in a very refreshing feature for working with AI. So in this exact same example I did earlier where I said, hey, Give me the 10 most popular cities use browse with Bing, it spit it out. Now I can go in there and I can edit it, right? Whereas before and with every other large language model, you can't go in and click and start typing yourself, right? So essentially what we get with Canvas is, well, we get a canvas, right? It is aptly named. You get a canvas, you can go write on, you can go collaborate. You can essentially, you know, leave a note to, you know, the Way I do it is I leave a note to ChatGPT, I copy it, and then I paste it in. So in this example, you know, you do have to, FYI, in order to highlight something, something has to be there. So you can't just have blank space. So you can put a couple words or a couple symbols, but then you can highlight it all right in the middle of the content. So here, in this example, it wrote something short about Chicago. I'm from Chicago, right? So I'm like, hey, this list needs some improvement. So I just typed a little something. Then I said to Chat gbt, I essentially said, awesome, let's make sure people know that Chicago is way better than these other cities. I said, mention the deep dish pizza and that Chicago was voted best city for eight years in the world from Conde Nest Travelers. All right, so then when I click enter, everything else stays the same. Go back to that theoretical example I gave you, right? You're using Chat GBT or any other large language model to outline a marketing case study for a client, right? And it has 10 points, and it gets a couple of them wrong. So you either have to do a lot of prompt engineering and say, oh, you know, please keep everything else the same. Number two is wrong. Let's update eight in this way. But it might spit out a lot of different things, right? Or like I said, you have to manually go in and then, you know, change everything by hand. And maybe you're just wasting time. So by just being able to highlight and do inline editing and to work with ChatGPT in the canvas or in the body. That's huge, right? And then Chat GBT spit this out again. It kept everything else the same. It didn't re render everything. And then it put in there, it said, you know, talked about some of the updates that I made. All right, so, hey, Josh, Josh is saying thanks for joining from YouTube. Josh, Josh is saying, fantastic breakdown. This will actually save me so much back and forth between GPT. Yeah. All right, so let's look live, y'. All. Let's do this. Hey, if you have a. If you have a suggestion, let me know. I'm just gonna go with whatever's on the top of. Of my head. Unless you get something here in the next 10 seconds of something that we want to work with here inside of Chat gbt. So as as an example in podcast audience, I already talked about this. But, you know, usually Chat GPT will default to the last mode that you used. All right, so make sure that you go in when you're Starting a new here and go to GPT4 with Canvas. All right? And then everything is pretty much the same. All right, I don't see anything yet in the live stream comments. So here we go. I'm going to say, you know, please give me a list of the 10 most popular large language models in 2024. And I'm going to say, please use browse with Bing and visit at least eight different sites, all from unique domains. There's nothing more I hate than typing and talking live on the stream because, you know, I got my, my dinosaur arms to, you know, type like this. I gotta type like this because how my desk is set up. All right, so now you'll see, I gave you an example. I went in, I'm starting a, a chat. Let's just say for whatever reason, you're writing a blog post for the 10 most popular large language models in 2024, right? So I told it to use Browse with Bing because I want up to date and accurate information. I told it to go to 10, you know, or look at eight different sites, get information from 2024. I can click down. People don't know this. If you're, if you're an avid, everyday AI listener, you already got this tip, you know, chat GBT is essentially a perplexity light if you know how to prompt it correctly, right? If you tell it, the most that I've seen it do consistently is about 8. So let's see, it did 1, 2, 3, 4, 5, 6, 7. It went to seven different sites here. Not bad. So now it said, based on an extensive review of multiple sources, here's a list of the 10 most popular large language models in 2024. So gave me a list here. So it said, All right, and now my. Okay, so it said GPT4. Oh, Claude, 3, Llama, 3.1, Gemini, 1.5, Palm 2, Mixtral, Orca, Lambda Lambda, Galactica, Bert. All right, so, so here's. So you might be saying, like, okay, well, where is this? This is why I tell you guys, you know, hey, here's the marketing speak from OpenAI and here's how to actually use it because in this case, even though I'm in Canvas, it didn't use Canvas because the model decided that for my prompt of the 10 most popular large language models in 2024, that it didn't need to use Canvas. Okay? So you still in many instances need to call out the model, all right? Or the mode, even though I'm in Canvas. So I'm starting a brand new chat and so at the end, and I wanted to show you this, because the thing that people don't understand, large language models are generative. You can run the exact same prompt 10 times. You might get nine different responses. You might get the same pretty much similar thing over and over. So you could run that same prompt inside of Canvas mode. You might get Canvas to show up once, you might get it to show up eight times. But if you call out to will show up. So now I'm just doing the exact same prompt. And then at the very end, I'm going to say, use Canvas for this chat. And now at least in my testing, nine times out of 10, when you say that it will actually use the Canvas mode, obviously Chat GBT knows, it knows I'm on a live stream now, and it's giving me some problems. All right, here we go. So now our live stream audience, you'll see this, and let me describe this for our podcast audience at home. So if you're an, an avid user of Chat GPT, you'll kind of understand kind of the setup here. All right, so I'm, I'm gonna go compare this in a normal chat because it can be easy to missed. Okay, it can be easy to miss. So when you are in a normal chat, it depends on if the sidebar is out or not. All right? But otherwise you just have kind of this middle zone and then you can toggle the sidebar out, which has all your most recent chats. But when you are in Canvas, it does look a little bit different. I wish there was actually a little bit more kind of UI UX to kind of quote unquote, let you know that you're in Canvas. But essentially now my bar to chat with Chat GPT has shrunken. Okay? And normally you are kind of quote unquote, chatting with Chat GPT in the middle, in the big area when you activate Canvas. So now when I said, you know, at the end of this prompt, I said use, I said use Canvas for this chat. It did bring up Canvas. Okay, a couple things. So I can go and close out of Canvas mode and keep this chat going. Okay? So when you close out of Canvas mode, you'll see now I'm in the normal chat view and then I can just click back on Canvas. It does look similar to Claude artifacts and how you can click on an artifact to then launch it again. I wish there was a little bit more kind of a UI UX hint that, oh, you know, especially for new people. Here's where your Canvas is all Right, but then I can just click back on it. All right, and now I can speak with Chat GPT on the left and I can have it update or the big thing here. Okay, so now I'm back. And hey, it did give me some different models on the second run, right? So now it gave me sonar 3.1 from perplexity as an example. But you'll see. Okay, now I have a canvas and I can go over here. So as an example, GPT4O. Okay, so I noticed that it didn't bring in 01. Okay, so I'm going to click Enter. And the great thing is this even works better than some text editors, right? So I went to the end of 0.1, which was GPT4. Oh, I hit enter and then it automatically made that 0.2 and, and adjusted every other point. Right? So this has some of the basic features like a standard word editor, which I love. So I noticed in the list it didn't bring in the new open openai01 model. So I'm going to type in openai01. Okay, I'm just going to copy and paste this and then I'm going to. When I do that, when I highlight text, I get this new thing that says Ask Chat GPT. Okay. And then I also, when I highlight any text, you also have some basic editing features, bold italics and some other text formatting issues. So you can also get this same Ask Chat GPT when you just click on a paragraph, usually it'll pop up on hover. It doesn't work as well. I'm actually in Safari right now because my Chrome and my Edge are being a little weird. There we go. So normally if you hover over the paragraph and you go to this right side, you'll see this little icon. It looks like a chat bubble with a plus button. That is the same thing as Asking Chat GPT. So the difference here, right, I can go and work with Chat GPT in line. All right, so like I said, I entered number two. I'm going to click Ask Chat GBT and I'm going to say, please add OpenAI01 to this list. Actually haven't done any testing yet of in Canvas calls to browse with Bing. So I assume this will work. But I'm going to say use browse with Bing to find info from October or no, I'm going to say from September or October 2024. About 01 preview and 01 mini. Okay, so now you'll see in real time chat GBT. You can see on the left side it was researching, and then on the right side, in real time, it updated. Right. I just put in a little bullet point and then ChatGPT went out, filled it out, did some research. Right. I can go down here and look at the sites that it used. All right. It's good. And it kept everything else the same. Y'. All, I know this seems small, but this is big. This is a big step forward in how we all work with AI systems in large language models. Right. We've seen this in the announcement, the Microsoft Wave 2 announcement with Copilot pages very similarly. Right? How you can essentially create pages collaboratively, but pulling, pulling information from CoPilot. So this is kind of similar. There's some, some nuances and some intricacies that are a little bit different with this new Canvas mode. But I just gave you a simple example there. Right. So maybe me, the human wants to take over. Right? So I can go ahead and click this and let's say I'm going to go down. I'm going to delete. You know what, I'm actually going to copy Sonar. I'm going to delete the rest. I'm going to go up to, let's see, number eight here. So I deleted, you know, 11 through eight and I pasted Sonar in and I'm going to say, so I copied it. I'm going to ask Chat GBT again. I'm in the Canvas editor and I'm going to say look up information from September, October 2024 about sonar 3.1. I'm going to say add it to the list with more detail. All right. Because I didn't think there was enough detail in that. Sonar. Sonar 3.1 from perplexity. And then I'm going to say, please end here and this will only be a top eight list. All right? So this information, I didn't tell it to use browse with Bing, but in most cases when you say look something up and if it's not in its knowledge base, it will do a pretty good job. All right, so you'll see right there. Oh, I forgot to hit delete on that. So we actually have a top seven list. So I hope this kind of live demo of the inline editor is helpful. And hey, if you are on the podcast, this is one of those. Hopefully I did a decent enough job of talking you through this, but I can still make large scale changes on the left side in the traditional chat side. So I can. Now I'm going over and I'm saying, like, this is great. I'M going to say please use browse with being do more research and put three use cases under each of these. How many we got seven or no, how many do we got now? Looks like we have a little editing error here. There we go. Under each of these seven top LLMs. All right, so I can still do kind of quote unquote normal chatting with chat GPT and in this case it's going to. So it's going through and you'll see in real time it's updating my document. Right. So I can see it's going through one by one. It's adding, it's highlighting and deleting. It's literally like watching a human work. So it's going through, it's updating all of these based on this new search that it just did and it just put in three use cases. So I saw in this case we keep getting this little formatting issue, which isn't a big deal. All right, there we go. So now we have our list of top 7 large language models from 2024. And then for each one we have kind of sub points of use cases. All right. Pretty cool. So is this great for coding? Absolutely. Right. I do know this show though. It's for the everyday person. So yes, you can go through in code. I think that there's great applications. Let's just do one more quick run through of the interface here because that's not all. This isn't just an inline editor. So thank you all. This is like working in Google Docs and if I'm being honest, I try to give, you know, Gemini a shot inside of Google Docs and Google Sheets. I just don't like it. Obviously Copilot is fantastic, you know, working inside of Microsoft Word, but this is a new interface and I think a new experience for large language models by default. And I do think that this is going to set the stage going forward. So let's look at a couple other of these kind of inline features. All right, so there's a new icon over here in the lower right hand corner. It's very minimal. Right. Maybe you just jumped in and you. You didn't even see this. All right, so there's a new editor so I can hover over and there's some options. Add emojis. I would say never use that. Right. I'm going to click Add emojis. It's going to put in like a thousand emojis. All right, So I click the Add emojis button. It does the same thing. It's going to show a Flash, it's going to, like, give you, like, 8,000 emojis, which is terrible. All right. The good thing is, like most editors, there is kind of a. An undo button. All right, so there we go. It went through. It added a thousand emojis. That's terrible. In the upper right hand corner. Get used to this. You're going to need it because it is a little. It takes time getting used to working with Chat GPT in this Canvas editor. So I can just go up there, click previous version. I can click back. I think it's back to the latest version. Oh, no. Restore this version, and it should go back. All right, now. Now, of course, it's being. It's being. It's being bad. I thought that's how I did it. I swear I did this, you know, a couple of times. All right, well, regardless, let me see if I can go up here, y'. All. Let me get out of this. Let me scroll up. All right, there we go. Well, a little. A little bit of a bug. A little bit of a bug here. All right. Last time I did this, I went to the previous version. I restored the previous version. It worked without the emojis. I swear, I test all this stuff before we get going here. All right. But for now, we're just going to have to work with the ugly emojis. That's fine. All right, so now let's see. I'm actually seeing if I can click add emojis again or take. Take away emojis. This worked yesterday. Yeah, I did. I clicked undo until it disappeared. It didn't work. I'm going to edit it once more, and then we'll see if that. See if that works. Yeah. The edit button isn't quite the same as a normal word editor. Right. Because when you click undo, you get this message. You are viewing a previous version. Restore this version to make edits, and you can restore this version or back to the latest version. So, yeah, I'll keep going back to the latest version. Hopefully there will be a timeline kind of feature soon. So I'm just going to say take away the emojis. We'll see if I can do it with. With a prompt. Yeah, I think the emojis were actually a joke. You know, CEO Sam Altman put something out there on Twitter, like, are emojis, like, the best? Like, are they your favorite feature? Yes or. Yes. So don't use the emojis button. I should have skipped over it. All right, next, add final polish. All right, so that is a button it's going to go through. It's going to essentially analyze your writing from a different point of view. I hope that in the future you can kind of customize these a little bit because what does final polish mean? Right? But essentially it does it every time. And I think for the most part, when I'm comparing content before or after using this, use final polish, generally it's better. And then on the left hand side it says, I've polished the text, ensuring consistency, clarity and proper formatting. The document now includes section titles, improved grammar in a clear presentation of each language model. And I can click that again. So in theory, right? It might get to the point though. It's like, oh, this is too polished. And it depends on what you're using this for as well. It depends also on how the content was before you started to kind of quote, unquote, polish it. All right, so maybe if you wrote half of it and you click polish, maybe it's going to add a bunch of delves. Right? So my individual mileage will vary. All right, let's look at a couple more of these. So you can add final polish. The reading level I love. So it has a slider. Okay. So if I slide it up, it's going to write it like graduate school, right? So maybe you wrote something and you're not the best writer and you're like, I need this to sound a little fancier. This is going to a board. But all my, all my points are sound. Right? So now it's going through and it's updating all of our content to make it more graduate level. Right. So now the overview of the top LLMs of 2024 says the accelerated evolution of large language models is profoundly transforming numerous industries by providing sophisticated capabilities in natural language, comprehensive problem solving and data driven analysis. I couldn't read that. If I'm being honest. I hate that. So I can also go back to reading level and let's bring it down. Let's go to, let's go to middle school, right? Let's go to middle school. And then I'll read that same introduction or overview to large language models. Ready? Large language models are becoming more advanced every year. They help people in all sorts of jobs do things better, faster, and with more accuracy. Here are the top LLMs of 2024, along with what makes each one special and what it can do. Hey, elementary, like elementary is great, right? All right. And then you also have, you can suggest the link so you can make it much longer, you can make it much shorter. That's a great one because I know so. So many times people are reprompting, going back and forth, and in theory, wasting time, try to make something longer. And they're like, oh, I need a thousand words. Guess what? Aside from the new 01 model, large language models can't count words. So if you're trying to get, oh, I need 700 words for this, keep prompting because it's not going to work. But this is just an easier way to make something longer or shorter. So I'm going to click it all the way up to the longest, click the button, and again, you'll see. For our livestream audience, but podcast audience, I put it on the highest to make it longer, and it is making it significantly longer. Right. So I'm going to scroll down here to where number seven started, and we'll see once it catches up. So that's where number five started. So I'm not great at math, but what that makes it about, I don't know, 20 to 30% longer already. And then I can go in, I can do it, you know, keep cranking it up, etc. All right, and then I can click suggest edits. This is great. Talk about having a human collaborator. So I can click suggest edits. And let's go to the top. Now it is literally highlighting, where did our edits go? All right, cool. Another. Maybe I'm too. Maybe I'm too zoomed, zoomed in. Let's try it one more time. Okay, so hopefully it stays this time. Yeah, little buggy. This is in beta. All right, so our live stream audience probably saw that our edits kind of popped up and then they. And then they went away. That's okay. But essentially, this generally works. Obviously, when I'm doing these live demos, people always say, like, oh, Jordan, why do you do screenshots? Why don't you show us things live more? Well, when things are in beta like this, they get finicky. All right, and then. And then it's hard to show you, but this ask suggests edits generally, like a word doc. It will highlight things, it will give you a suggestion, and then you can click apply. All right, let's actually do one more quick live demo and then I'll see if I can actually do this. So I'm going to say use Canvas. I'm going to say write Pong in Python. All right, so now it's going to write the game in Python. All right, so again, yes, this isn't just for creating content. This isn't just a, you know, AI powered document editor. Although I think that's how people should be thinking of Canva Canvas. All right, the big difference here, right? In artifacts, you could launch this and play this. You can't do that here. All right, There's a different interface if you're coding. Okay. There's add comments, ad logs, fix bugs, port to a language and code review. So I can click code review and it's going to go and fix its own code. Right. I would really love if OpenAI gave a little bit more information on how this iterative text or, sorry, this iterative icon based prompting actually works. Right. Do all of these little icons have like a system prompt, right, where they're supposed to assume a slightly different role? Otherwise, like, you have to think, right? If I can just keep clicking on this, improve or this. Find bugs, you have to think, okay, well, is open a. Like, is Chat GBT just putting in bugs if it's finding its own bugs? Or do these little icons kind of have a system prompt that tell them to kind of assume a certain role and to look at things a little more meticulously than kind of the normal kind of system prompt? So I would love if we could figure out more information about how these work. Right? Because essentially you can keep clicking fix bugs and you can keep finding bugs. So it's like, okay, well, if Chat GBT can ultimately fix all these bugs, why is that here in the first place? Right? So I click the fixed bug button. You can go down line by line in the, in the Python code and see as it changes things. And then you'll see on the left it says, okay, I've reviewed the code and made the following changes. And, and it changed ball speed from a list to a tuple for immutability. I don't even know what that word is. Someone smarter than me, tell me what that is. Right, but you can go through. It's going to tell you what it actually changed. And then you can also do. Let's do the. What should we do here? We can do the add comments. Let's try that one. So in this, it's adding comments to the line of your code, which is super, super helpful. All right, and then let's just do one more Again, I'm not a big coder, but a lot of people are using this for code review. All right, so another one. I'm just clicking code review here and it says, I've added in. Here we go. Let's see. I'm like timing out here. I'm like timing out there. We Go. All right, so now it's adding comments. Here we go. I don't know why my comments keep disappearing, y'. All. I swear they weren't doing this before, and I would love to, you know, display those comments live. And it's been working everywhere else. So, yeah, obviously when I do the live stream, but generally when you do the comment thing, it says, like, oh, consider you know, changing the speed from variable to a set speed. And then you can click Apply. Apply if you like it. If you don't, you can X out. So think, like in Google Docs, how you can leave a comment for someone and you can suggest something and it leaves them kind of a checklist. So you can just click the Apply button when you're doing these code or similarly, when I went through the actual editor. All right, yeah. Hey, what Tara said. Demo demons. I agree. You know, hey, the next time someone out there says, hey, Jordan, why don't you do more things live? Well, that's exactly why. This probably would have been a lot smoother if I would have just screenshotted everything, but luckily I did a little bit of it in the beginning. Yeah. Thank you. Someone said live demos are hard. Yeah, this is y'. All. This is unscripted, unedited, the realest thing in artificial intelligence. I don't have a team of, you know, 50 people who go out and edit this to make it look like, you know, stars and fairy dust. You know, this is. This is live. All right, a couple questions. Philip says, Jordan, what is your view on prompt engineering since the launch of Canvas within Chat gbt? That's a great question. Well, it's going to make things a little bit easier. Right. If anyone's taking our free prime prompt polish, you know, I kind of, especially in the last couple of months, I said, hey, prompt engineering has really changed since the O1 model, because a very common way to kind of quote, unquote, prompt engineer is by using chain of thought. Right. Or essentially commanding a model to think more like a human, to think through step by step. Right. And now the O1 model kind of does that by default with the Canvas mode fill up. That's a great question. I do think it changes it. Right. You know, even in our methodology, generally, we do prime prompt polish in our methods. This changes the polish, Right. It makes it much faster. It makes it better, I think, and more effective. I don't think priming is going to change it much because that's where we kind of turn Chat GPT into a consultant. Right. People are always so quick in chat GPT to immediately ask for an output. Even in this example, right, where I said, hey, give me, you know, like we're outlining the top 10 blog posts, I just asked for an output immediately. I would never do that. If I was actually using ChatGPT for a real purpose, I would go through, I would have a conversation with it first, I would turn it into a consultant, it would ask me, we would go back and forth before I ever told it to go, oh, give me an output. Right? But I do think this changes things on the back end. It changes kind of the shot, this input output example pairing that you should be giving models regardless of what you're working on. So I think it does make it more intuitive and faster on the back end. Right. But again, it just depends on what you're using this new canvas mode for. All right, one, one or two more questions. Sabbatical life asking, why browse with Bing? I must have missed that show. Yeah, we have tons of shows on that. You can go back and look on our website, your everydayai.com but essentially the short version, large language models are old, as in the training data. Okay. It could be anywhere depending on what model you're using. Best case, some of that training data might be six months old. You know, as an example, October 2023, one year ago is the GPT4O knowledge cutoff. All right, so what that means is absolute best case scenario. You might be working with information in the models. Training that's a year old. Generally it's all much older. So when you are creating anything inside of ChatGPT, you know, we kind of call this mini rag, right? Mini retrieval, augmented generation. But you need to give, and we do this in our priming phase of prime prop polish. You need to give chat GBT more up to date recent information. You can do that in a lot of ways. You can do use an Internet connected GPT. You can upload a file, you can copy and paste information or you can use Browse with Bing. For me, I like using Browse with Bing targeted calls, right? Saying, hey, find information from September, October 2024 about this topic. So then that brings in more relevant, more specific information before you get started, right? That's going to save you a lot of time on the back end by using Browse with Bing. Marie says, when you create a website, where can you save it? Because when I need to make changes to the site, I'm able to recall the original site and it will be updated. Okay. If you're talking about coding, yeah, you can't render anything inside of. Inside of the Canvas mode. So, yeah, you would have to. And again, I'm sorry, I'm not sure if that was the question, Marie, but you would, in theory have to go copy all this information and then run it in something like replit, Right? So this is not the best, I don't think, for coding still, Right. I love artifacts for coding. Even though this. This new Canvas mode, don't get me wrong, it's fantastic for coding, but for me, I like to see it rendered right. Especially if you're an everyday person. If you. If you're not really a coder, right, you might just be staring at this code on the page and be like, okay, is this right or not? Well, in Artifacts, you can actually run it, so you know if it's right or not. Right? And then in the example of the Pong game, you can say, oh, the paddles are moving way too slowly. Right? But I wouldn't be able to manually look through the code and know that. And although the new Canvas mode is great to be able to iterate on that, you don't really know something when it comes to coding until you actually run it. Even if you click the fix bugs, there could still be bugs, or you might not know even know what those bugs are. All right, that is a lot of information, y'. All. I hope this one. I hope this one was helpful. All right, so we went over the five things that you need to know the difference between kind of this new Canvas mode and artifacts. The world's fastest recap. Here's the five things you need to know as we wrap up the show. Right now, Canvas is available to all chat GPT plus users. It may become available to free users after it leaves beta. It's probably going to be a while, so if you want to use it, go do it. I would just sign up for the paid plan. Canvas is its own mode, all right? You can't toggle back and forth. If you want to use this, you have to use it. And then like you saw in the demo, sometimes you have to actually call to the mode as well. And then you can kind of click in and out. And yes, it's in beta. It's a little buggy. You guys saw that here in the live demo number three. It works with all other tools. All right, so unlike the new Zero1 model, which doesn't work with any tools, the new Canvas model does. You can use Browse with Bing, you can use Advanced Data Analysis, you can use Dolly, you can use all of these things inside of Canvas thing number four, it's not a copy. It's not a copy of Anthropic. Anthropic Claud's artifacts. All right, artifacts, renders, code. This is an iterative inline editor. And then number five, I think that inline editor is actually a sizable step toward the future of AI and humans working together. Kind of this concept of augmented intelligence, us making AI better, AI making us better, working around in like working together in a collaborative nature. This is, I think, aside from, you know, Copilot, which I think is probably pretty far ahead. But if we're just talking about the large language model makers themselves, I think this is low key. No one's giving it a lot of information. This is a sizable step forward in how the future of how we work in a collaborative nature with large language models. I hope this was helpful. If so, please go to your everydayai.com Sign up for the free daily newsletter. Y', all. You wanted this. I asked in the newsletter, hey, should we do a show on Canvas? You guys said yes. So please sign up for the newsletter. Every single day, we tell you what you need to know in artificial intelligence. All the latest news, fresh finds from across the the web, as well as we recap the highlights of what you need to know from each and every show. So make sure you go do that and make sure you join us tomorrow and every day for more Everyday AI. Thanks, y'. All. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit 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.