Canvas is a new feature by OpenAI that enhances ChatGPT for writing and coding projects. It provides a side panel for real-time editing, suggesting edits, adjusting text length, changing reading levels, adding polish, and even inserting emojis. For coding, it offers code review, bug fixes, adding logs, and porting code to different languages.
Canvas improves writing tasks by allowing users to edit text in real-time, suggest edits, adjust text length, change reading levels from kindergarten to graduate school, add final polish for grammar and clarity, and insert emojis for emphasis. It eliminates the need for repetitive prompts by enabling targeted edits within a document.
Canvas offers coding features like code review, bug fixes, adding print statements for debugging, inserting comments for better understanding, and porting code to different programming languages such as JavaScript, Python, and C++. It allows developers to select specific lines of code for targeted improvements.
Canvas is designed to automatically trigger for tasks like writing blogs, making targeted edits, or rewriting documents. OpenAI trained GPT-4.0 to recognize when to open the Canvas panel, avoiding over-triggering for general queries like recipe requests. Users can also manually trigger it by clicking a button or adding 'use Canvas' to their prompt.
A key challenge was defining when to trigger the Canvas panel. OpenAI aimed to avoid over-triggering for general queries while ensuring it activates for relevant tasks like writing or coding. They conducted over 20 internal evaluations to improve correct triggers, achieving an 83% success rate compared to a baseline GPT-4 model.
Canvas is seen as a more advanced version compared to Anthropic's Artifact. It offers robust features for both writing and coding, including real-time editing, code review, bug fixes, and language porting. Artifact, while similar, appears to be a more limited or 'nerfed' version in comparison.
Canvas could disrupt startups, especially those focused on code editors or writing tools. Developers fear that features like multi-file edits and testing environments could render their apps obsolete. The launch has sparked concerns about the survival of startups in the AI ecosystem.
OpenAI has just launched a new feature, which they are calling canvas. I'm going to break down everything that it's able to do. This is really good for anyone that uses chat GPT for writing or developers using it for coding. They have a bunch of new features and this is completely changing the way that we are using chat GPT. They just posted over on, uh,
They said we're rolling out an early version of Canvas, a new way to work with ChatGPT on writing and coding projects that go beyond simple chat. Starting today, Plus and Teams users can try it by selecting GPT-4.0 with Canvas in the model picker. So right on ChatGPT where you would typically go and hit the dropdown to change what model you're talking with, it's going to appear right there. You're going to be able to go grab it.
Uh, right there. So what it looks like is just this little pop-up module on the side. It's going to be like a little sparkling pencil over on the right hand side. When you open this up, it's
It's going to open up an entire new screen or panel that you're actually going to start being able to edit what you're working on. There's a bunch of new features, a bunch of new buttons that are really interesting. So let's get into them. Specifically, I'll talk about anyone using this for writing first. I think this is a really amazing new use case. And the idea with this is that you're able to essentially better collaborate and write and something.
things that might have been prompts in the past that you use ChatGPT for, they're now all going to have shortcuts. The first one is to suggest edits. You can ask ChatGPT to suggest an edit. And what it's going to be, right, it pulls open this whole side panel. You'll be chatting with ChatGPT. It's giving you your answers. It opens this full-on side panel where it's kind of like a document editor, like Google Docs or something else. You select a piece of the text.
And you can then get it to manipulate that text in particular. So they have examples where they're highlighting the title of an article that you got it to generate and you highlight the article title. There's a little pop-up that comes up and you type in it, make this more creative and it's rewrites the title to be more creative instead of having to get chat GPT to redo whatever your thing is every single time and try to get one shot, everything perfect, right? With this really long elaborate prompt.
You're essentially editing and working in real time. They do some really cool things. They had a demo where you uploaded a document of, you know, a bunch of information that you'd saved. And you said, they said, hey, like take all this info and write an article about X, Y, and Z. I was able to do that. They then went and selected the introduction to the article. And they said, and they can make a bunch of different things.
a bunch of different edits to that. So one of them is adjusting the length. Sometimes you get chat to write something and it's just like, you know, a sentence and you're like, geez, like flush this out, make this a paragraph. Or sometimes it writes you three paragraphs and you're like, no, please. I wanted two sentences on this. So it just has a button that you'll be able to select what you're talking about and get it to redo that. And it's a nice, because sometimes I get it, like sometimes it'll write
a whole chunk of whatever I wanted correctly, but the beginning part is just not good. So you're going to be able to just like edit the little pieces of a document that you want. This is amazing for me.
The other thing that it's able to do is change the reading level. So you can adjust it from kindergarten to graduate school. There's a reading level that you're actually able to switch right inside of it, which is going to be awesome, right? Sometimes I'm like, okay, this is way too complex. I don't even understand, especially when it's really technical stuff. I'm like, I don't even understand this. Please just like explain it like it's going to a fifth grader and then I can understand complex topics better. This is going to be something everyone can do very simple.
The other thing is you can add final polish. So they just have a button to add final polish. You can do it to the whole document, to a specific area. It's going to check for things like grammar, clarity, and consistency, which is great sometimes if you're manually editing stuff, right? Like I find a lot of times I get it to write something, then I go and like change something. You know, I don't like this paragraph. I delete it. And then by the end of it, I'm like, is this whole thing like really consistent? Is my writing style the same as it's like, how do we kind of like fix that? So that's what this final polish button is going to do.
The last thing that they can do, they have a button specifically to add emojis, any relevant emojis. They say you can add emphasis and color. Probably not something I will use a lot because I'm always telling it to please never use any emojis, but that's just me. So maybe there's other people that are a little bit different. So that's some of the writing features that are absolutely fascinating and I'll show you
that's actually looking like in a second. I want us to talk about what, even if you're not a developer, but if you are, this will be interesting, but, um, what they're doing with coding and canvas. So what was previously done was anthropic, the biggest competitor to chat right now with their cloud model, they had something called artifact, which is similar, but it honestly seems like a bit of a nerfed version after seeing what the coding function can do here. Um, and you're able to essentially have a side panel that would help you with, um,
Um, your code. So the new coding shortcuts that they're adding with their, with their new canvas, number one is to review the code. So ChatGPT can give some suggestions to improve your code. Again, you can select a certain line of code and have it work on that or the whole thing. It can add a log. So it can insert print statements to help you debug and understand your code.
So that's very, very useful. And then it can add comments. This is something that developers on my projects that I'm working on. They use this all the time. Sometimes when someone writes the code and a new person comes in, it's hard to understand. It's hard. You gotta go figure the whole thing out. And something that my developers have been doing when code gets switched from one developer to another is they'll use chat.gbt to go and just comment the entire code, explaining how everything works, um, makes it so much easier to understand. And you just way faster than trying to read the whole thing. So.
The next thing it can do is it can fix bugs. This is really useful. Um, obviously code is not perfect, especially when it gets spit out of a different AI model. And so just the ability to have this thing go through and fix any bugs, um, that are in it is awesome. And essentially I've seen like the demos of this, it looks at every single line of code and it goes and it verifies that everything is running smoothly on those, which is fantastic.
Um, the last thing is that you can actually port this to a new language, right? Sometimes people will be like, Hey, like write me a code that can do X, Y, and Z and it writes it in JavaScript. And you're like, Oh no, I need this thing to be in Python. So you have a very simple way to change the actual coding language from JavaScript to TypeScript, to Python, to C plus plus a PHP, like whatever you're actually trying to work in. So this is very, very interesting.
How does this thing actually work? How does this panel get triggered? Is it manual? I believe there's going to be a manual button right on the side that you can click on. Um, it should be on chat GPT, uh, by the time you're watching, this should be right in the bottom right-hand corner of chat GPT. You click on that to open up a panel, but alternately they can, you can add to your prompt that you're saying use canvas. So, um,
This is kind of awesome if you're like, "Okay, I know I'm going to be needing a canvas." Now, they tried to build in some automatic triggers to make this thing pop up. They trained GPT-40 to essentially know when to make this pop up. The thing that they tried to get us to do is to trigger the canvas for writing and coding, generating a bunch of different content types, making targeted edits, rewriting documents, and providing inline critique.
If you're working on any of those things, they're going to try to trigger the canvas to open up in the side. So the problem is this could be annoying if you're just trying to get this thing to like, you know, write a recipe or, you know, help me with a recipe. You don't need to edit this thing or work on this thing. You're not, you just want it to give you the information. You don't want the whole canvas to pop up.
So they said a key challenge was defining when to trigger a canvas. We taught the model to open a canvas for prompts, like write a blog about the history of coffee beans while avoiding over triggering for general Q and A's like help me cook a new recipe for dinner. So they tried to really work on this. Um, and they had, I believe they met, they measured progress with over 20 automated in, uh, internal evaluations to try to get this working correctly. Um,
Um, they said for writing tasks, we prioritize improving correct triggers, um, at the expense of non-correct triggers. So it's about 83% compared to a baseline zero shot GPT-4 with prompted instructions. It's fairly good. It's not perfect. Uh, it's getting there, but honestly.
If you need it, you can click on the button that's going to automatically trigger. I think it's better that it under triggers when you can click the button if you need it versus over triggers. So everything that you don't want it to open, it's going to be opening. Okay. All of that out of the way. I think this is absolutely a fascinating new use case. And people are talking about this all over Twitter. With the big introduction that they gave, people are already, OpenAI developers, the official...
their official account over on X is showing a demo of what this would actually look like in the wild. And I thought it was sort of fascinating because essentially they're able to, you know, someone's asking it to, I have a 3d model and I want it to display in a 3d scene. I have a lot of two longitude. I would like to translate this into X, Y, Z coordinates. Can you write the code for me? Right? So it's, they essentially, um, are saying this now nowhere in there. Does it say to open up, uh,
the specific side panel, but it understands that this is what it needs. So it opens up, uh, it opens up the canvas now while they're, um, while they're doing this, they simply click on the button on the right hand side and they want to pour it into a new language. So there's a little adjuster where they say, switch this to JavaScript immediately. It adjusts the entire thing into JavaScript.
which is super, super useful. Then they're like, okay, this is cool, but I need to do like a specific, uh, you know, geolocation to earth that I want to, um, keep it at. So they go and add a little bit of code. And as soon as the code is edited now, not everything's perfect. So they're going to just go ahead and select the chunk of code afterwards. They just ask chat GPT. I want this to be relative to earth and boom, it takes the, uh, that code and it
It goes through the entire document line by line of code to check, to make sure everything's good. It rewrites the piece in question. Now I think the best feature of all of this, um, that I'm really excited for is the fact that.
it is essentially able to select one line of code or one line of your document and go and just edit that specifically. So overall, very, very excited for that. The responses on X have been quite positive. Someone said, interesting, exactly what I use cursor for, any specific differences, why I should use this feature. I would say ChaiGPT is probably just going to keep
developing and developing this feature. It's kind of rough for cursor this week. We had a whole bunch of, um, we had a whole bunch of like VS code fork projects. One in particular got into Y Combinator, got $500,000 and everyone was roasting them because they just copied an open source library. And there's a lot of memes right now that those code editors are really sweating because this is
this is obviously going to kill a lot of startups. So even as I'm scrolling down through the comments, I'm seeing someone that says if they add multi-file edits and testing environments, my app and 50 other startups are cooked. So
Again, it's kind of rough every time OpenAI comes out with a new feature. A lot of people are going to complain about, you know, the repercussions on the startup ecosystem because it does impact a lot of people working on these types of projects. I'll keep you up to date on everything else rolling out. This is an amazing new feature. I'm really excited to
Uh, try and to get started with, let me know what you think in the comments. If you liked the video, let me know. And if you're interested in making money online with AI tools, I would love to have you join my AI hustle school community where I do an exclusive piece of content. Every single week I make a video showing exactly what AI tools I'm using and the side hustles that I do. So that sounds interesting to you. You can check out the link in the description to my AI hustle school community. Hope you have an amazing rest of your day.