ChatGPT recently launched an innovative feature called Canvas, designed to enhance collaboration and editing capabilities. Upon inputting a prompt, users can receive feedback in a separate window, allowing specific parts of text or code to be edited, commented on, or updated. This new addition is particularly useful for writing and coding, featuring functions like editing suggestions for writing pieces and comprehensive code reviews for programming tasks. The Canvas is a part of the GPT-4.0 model, which has been trained specifically to utilize this tool, providing targeted edits without altering entire outputs.

The introduction of Canvas poses an interesting scenario for companies reliant on ChatGPT's platform. While OpenAI has established itself as a platform provider, offering tools for developers to create advanced AI applications, the launch of Canvas suggests potential competition with AI-driven Integrated Development Environments (IDEs) and coding assistants. Companies building on similar APIs are warned about platform risk—a challenge faced if the platform owner begins to compete directly, changing policies, or using collected data to outmaneuver existing businesses.

Main takeaways from the video:

Please remember to turn on the CC button to view the subtitles.

Key Vocabularies and Common Phrases:

1. platform risk [ˈplætˌfɔrm rɪsk] - (noun) - The risk associated with building a business on a third-party platform that could change its policies or become a competitor. - Synonyms: (system risk, dependency risk, infrastructural risk)

If you build your entire business on top of somebody else's API, you are at platform risk.

2. collaboration [kəˌlæbəˈreɪʃən] - (noun) - The action of working with someone to produce or create something. - Synonyms: (cooperation, partnership, teamwork)

The way to think about canvas is a new way to collaborate with chat GPT.

3. edit [ˈɛdɪt] - (verb) - To make changes to text or code. - Synonyms: (modify, revise, amend)

You can ask chat GPT to edit certain parts of text or code.

4. iterative [ˈɪtəˌreɪtɪv] - (adjective) - Relating to a process that is repeated several times to improve the outcome. - Synonyms: (repeated, cyclic, sequential)

It looks like maybe it's thinking iterative improvements to the game is the same as a bug fix.

5. synthesize [ˈsɪnθəˌsaɪz] - (verb) - To combine different elements to form a coherent whole. - Synonyms: (merge, integrate, blend)

This is a tried and true model for platform companies to synthesize new successful businesses.

6. port [pɔrt] - (verb) - To transfer software to a different system or environment. - Synonyms: (transfer, migrate, convert)

For coding, it reviews code, it can add logs, add comments, fix bugs and port to a different language, which is pretty cool

7. infrastructure [ˈɪnfrəˌstrʌkʧər] - (noun) - The fundamental facilities and systems necessary for an entity to function effectively. - Synonyms: (framework, foundation, structure)

And this feels very analogous to what I went through at my previous company. We built a lot of our infrastructure on top of a company called Twilio.

8. sophisticated [səˈfɪstəˌkeɪtɪd] - (adjective) - Having a refined knowledge or complexity often used to describe methods or technology. - Synonyms: (advanced, refined, complex)

Yeah, so they're definitely using longer, more sophisticated words.

9. utility [juˈtɪlɪˌti] - (noun) - The state of being useful, profitable, or beneficial. - Synonyms: (usefulness, advantage, benefit)

This feature feels akin to Claude's artifacts and it could do a whole bunch of stuff.

10. autonomous [ɔˈtɒnəməs] - (adjective) - Freedom from external control or influence; independent. - Synonyms: (independent, self-governing, self-sufficient)

This is actually a different version of GPT-4 O that recognizes when it should use canvas.

Huge ChatGPT Upgrade - Introducing “Canvas”

Just yesterday, ChatGPt launched a brand new, pretty incredible feature called canvas. This feature feels akin to Claude's artifacts and it could do a whole bunch of stuff. So that's what we're going to go over today. The way to think about canvas is a new way to collaborate with chat GPT. You type a prompt and instead of it just being in one single chat, it opens up this separate window where you can ask chat GPT to edit certain parts of text or code. You can ask it to update, you can ask it to add comments. There's a whole bunch of cool stuff. So I'm briefly going to tell you about it, then we're just going to play around with it.

So everybody already has access to chat GPT with canvas. And what I really like about it is the fact that they didn't say it's coming soon, they just released it the day that they announced it. So if you click in the dropdown, here it is, GPT four o with canvas beta. And they said for now, everybody with a pro or enterprise account will get access to it, and then as soon as it leaves beta, every free user will also get access to it.

There are two main use cases that canvas looks to help with writing and coding. And obviously I definitely want to test out the coding aspects, which we will do right after this. One thing I found really interesting is they actually trained GPT four o to collaborate. So this is not just the base GPT four o model with some code wrapper around it. This is actually a different version of GPT four O that recognizes when it should use canvas.

So for writing, it includes suggesting edits, adjusting the length, changing the reading level, adding final polish and adding emojis. For coding, it reviews code, it can add logs, add comments, fix bugs and port to a different language, which is pretty cool. And as I mentioned, GPT four o was trained to use canvas. So to do that, they developed these core behaviors triggering the canvas for writing encoding. So that's obviously just getting it going, generating diverse content types, making targeted edits, which is super interesting.

So rather than redoing the entire output every time, you can actually just select a certain piece of the content and it will update that only. All right, so now I just want to test it out and show you what it's like to use it. But before I do that, I actually want to speak to something that I think a lot of companies that are built on chat GPT should be nervous about.

Now, OpenAI, as a company, has always described itself as a platform company. They are going to give other developers the tools that they need to build amazing AI applications. But if I'm an AI company and I'm seeing this, I would be a little bit nervous. And this feels very analogous to what I went through at my previous company.

So at my previous company, we built a lot of our infrastructure on top of a company called Twilio. Twilio provides telephony services and they were truly a platform company. All they did was provide an API until they didn't. And we built an entire suite of software on top of Twilio using SMS and phone calls and other things.

And then one day Twilio started competing with us. They started building on top of their own platform. And so if you've watched this channel at all, you know, I've talked about platform risk in the past. If you build your entire business on top of somebody else's API, you are at platform risk. If they decide to change their policies, increase their price, compete directly with you because they saw a really successful business model, they will do that.

There is nothing actually preventing them from doing that. And in fact, they get all of this incredible data, they get to see what all of these companies build on top of their platform and then they can cherry pick the best ones and recreate it. This is a tried and true model for platform companies. And so now with canvas, I'm looking at continue and other AI based ides, and they should be nervous right now because this is essentially the first step in the direction to compete with those companies.

All right, so enough of that. Let me know what you think of platform risk building on top of OpenAI in the comments. All right, so we have it GPT 4.0 with canvas. Let's just play around with it first. Of course. Let's get it to write the snake game.

Now, one thing that canvas doesn't do is execute code, and I believe artifacts does do that. So that's one thing that it's missing. All right, so here it is. So it popped open a separate little area of the window and it looks like a code editor. We can delete stuff, we can add it back easily.

Here's the actual conversation over here. But it removed a big part of what would have been there, which is the code, and place it in the separate window, which I love. So we have the game snake, we can't execute it. Plus that's not the point of this video. So I'm not going to do that. But let's see, rather than these colors. So we simply highlight then down in the right corner we have all of the different options. So add comments, add logs, fix bugs, port to a different language and code review.

So instead of doing that, I'm just going to highlight it, click ask chat GPT, and I'm going to say change these colors to something else. So it should. Yep, look at that. That's really cool.

So it just changed these colors. Everything else stayed the same. You see the little editing update right here in the actual chat window? And that's it. I've updated the colors to new shades for better variation.

All right, so that's how the basics work. So what I could also do is highlight pygame and ask it to use a different library, which would really change the entire application. So use a different library. And I love so now it's using turtle and it actually has to change a lot of things in the code and it knew to do that, which is fantastic. And I love the little animation that it has where it's just highlighting the code as it's changing it.

So I've updated the code to use the turtle library inside of Pygame. Let me know if you need further modifications. All right, so that's what that looks like. But there's not many comments in this code right now.

So let's ask it to add comments. So first let's ask it to add comments to this set of lines of code right there. So I'm simply going to come over here and I'm going to click add comments. And actually I was wrong. It automatically adds comments to the entire document, which is kind of weird because I only highlighted these, but now it's adding a ton of comments. But let's say we don't like this. Let's say we don't like Python. So let's port it to a different language.

So I'm going to click port to different language. And look, we have a slider php, c Python. Keep current JavaScript, typescript Java. Let's do JavaScript. So I simply select that and it should rewrite it in JavaScript. So easy. So it's done. Now it's in JavaScript. Now let's try porting it to something like C.

Yep, this looks like C. All right. So I'm actually going to switch it back because I don't know c very well at all. I'm going to switch it back to Python. Now let's add some bugs. So right here for the move snake, I'm going to use up twice and let's see if it's able to catch that. So I go over here again and I look for the fix bugs icon. I click that and let's see what it says.

So it's interesting, it's actually changing a lot of the code already, which I don't know why it would have to do that to fix the bugs because there was really only one bug. Let's see if it's able to fix that one. And it did. Okay.

The down movement direction was incorrectly assigned to up. Now, something interesting is, as I mentioned, it changed a lot of the code, which is not exactly what I wanted, but that's fine.

Let's see what happens if we just do fix bug one more time. Let's see if it just keeps finding bugs and maybe the way it was programmed, it will just always find a bug with it. Yeah, so it's definitely changing the code again just a little bit. But it is changing it, especially down here in the main game loop. It is changing it quite a bit.

I reviewed the code fix a bug where the game did not reset properly upon wall collision. Okay, so it looks like maybe it's thinking iterative improvements to the game is the same as a bug fix. That's okay. Alright, now let's add some logs.

First, let's just try to add logs to something that I highlight. I wonder if it's going to do that. Nope, it's adding logs to everywhere, so it doesn't seem like just highlighting the code does anything unless you are using that as GPT feature. So it added a bunch of logs. Now let's see if I can remove the logs. Now remove the logs.

Okay. Yeah, so it is removing the logs. All right. I mean, this seems pretty cool. I think this is very basic to start with, but I like it and it's definitely much better to kind of go through the iteration process of the code here and then paste it into versus code. But honestly, this is kind of what continue and other ide AI coders already do.

So they still have a long ways to go, but it seems clear that they are definitely heading in the direction of competing with AI coding assistance. All right, now let's try some creative writing and let's see what it does. So I'm going to start a new chat. So write me a story about strawberries becoming intelligent intelligence. I miswrote that. Okay, so we have a story and just like I expected, it should have output a new canvas area of the page.

So does this story capture the essence of what you imagine? Let me know if you'd like to make any changes. So just like normal, we can come in here, we could delete stuff, we could add, remove. This is fully editable. And just like normal, you can highlight certain areas and do ask chat GPT just like that.

But we're going to wait on that. Let's go down to the options we have. So suggest edits, adjust the length, reading level, add final polish and add emojis. So let's first do suggest edits. This is really cool. Okay, so it's going to highlight different areas and kind of similar into Google Docs. It gives you little comments that you can actually accept or decline.

So here, consider adding more vivid description of the setting to draw the reader in more effectively and create a stronger atmosphere. So let's do that. So it should rewrite that part. There we go. Really good. Let's do it again right here.

So this says the nanobots were designed to enhance neural connections in living organisms and they worked their way into the strawberries nesting in amongst the seeds that dotted their red skin. This sentence could benefit from a bit more explanation about why the nanobots were designed this way. Okay, let's try it. Alright. Now it says, originally intended for medical research to improve brain function, so it gives more background story. So that's great.

So I'm going to decline these other ones. And now let's adjust the length. And again we have another slider. So I want it to be shorter and it should rewrite the whole thing. It's really cool, the little subtle animations that they have. So here we go, much shorter.

Let's see if we go even shorter than that. So I'm going to go shortest. All right. And what happens if we do it again? Let's find out. So I'm going to adjust the length again and go shortest again.

I wonder if it's going to go shorter or just keep it at the same length. Okay, so it did go a little shorter. Let's try to go shorter again. Just want to see where the limits of the functionality are. Yeah, so it'll keep actually just shortening it.

So now let's go longer. Okay, great. Let's try something else right now. I don't know the current reading level, but it's going to keep it there. Let's go to the kindergarten level.

So it's going to rewrite everything and just use simpler words, easier grammar to read, etcetera. So one day, special rain fell on some strawberries. Yeah, so this looks very good. Now let's change the reading level to graduate school. Yeah, so they're definitely using longer, more sophisticated words.

Definitely looks like this would not be possible to read by a kindergartner. So that's really cool. Now let's go back to a kindergarten level and let's see what happens if we select kindergarten level after selecting kindergarten level.

So there we go. And let's select kindergarten level again. Yeah, it looks about the same. So it just kind of rewrote. It didn't necessarily make it even more basic and that would make sense.

So let's go back and let's do high school level. All right, there we go. Now let's add emojis. Now, funny thing about the ad emojis feature, it's pretty much useless. I mean, unless that is exactly what you're looking for it to do. But why would you do that? And sam Altman posted a funny poll on Twitter yesterday.

Is adamojis and canvas the best feature OpenAI has ever shipped? Yes and yes. And so of course the answer is yes. So there we go. It added emojis.

Now I'm going to say remove the emojis over here. Oh, I see a lot of errors actually. So here's rain just out of nowhere. I think I saw a misspelling somewhere in here, but I'm not seeing it now. So maybe it corrected it.

And let's go to add final polish now hopefully it corrects this rain at the beginning it did. Okay, good. And it added a title now. So this is really cool.

So I've come to ada core.com and I'm going to give it some problems. I'm going to have it fix bugs. So I'm going to highlight the code that they have. I'm going to copy it, and the problem is uninitialized variables and fields.

So let's switch back to canvas and I'm going to say fix bugs in this using canvas. I don't know if I actually have to do that. We'll see. Yep. Okay, so it did open it up. Contacts, bug fix. So let's see. It's writing the code out first.

Good. And it didn't actually tell me what the bugs were. Here we go. I've addressed a few issues in the original code initialization of done in the complete procedure done was not being explicitly initialized to false null contact definition and get by name initialization. So these seem like what needed to be fixed. So I think it did catch this c dot name length on line, 86 the contacts complete, not initialize.

Let's look. Yeah, the complete procedure, I think that fixed it. Obviously I'm not checking exactly, but it seems like it definitely identified it. So that's the new feature chat GPT with canvas. Looks very cool, still very basic, but I think I if they continue to iterate, they're going to be competing with a lot of other AI products built on top of them.

Either way, let me know what you think of canvas. Let me know what you think of platform risk of being a company built on top of OpenAI's API.

Artificial Intelligence, Technology, Innovation, Chatgpt, Coding Assistance, Platform Risk, Ai, Llm, Artificial Intelligence, Large Language Model, Openai, Mistral, Chatgpt, Ai News, Claude, Anthropic, Apple Ai, Apple Intelligence, Llama, Meta Ai, Google Ai, Chatgpt Canvas, Matthew Berman