ChatGPT의 캔버스: 새로운 협업 글쓰기 및 코딩 도구

### ChatGPT의 새로운 캔버스 기능: 협업의 판도를 바꾸는 기능
* 🆕 이제 막 출시된 ChatGPT의 캔버스 기능은 **사용자가 텍스트나 코드를 편집할 수 있는 별도의 창**을 열어 보다 효과적으로 협업할 수 있도록 하여 전반적인 상호작용 경험을 향상시킵니다.
* 💻 이 기능은 현재 **프로 또는 엔터프라이즈 계정을 가진 모든 사용자**가 사용할 수 있으며, 베타 버전이 종료되면 **무료 사용자에게도 확대할 계획**입니다.
* ✍ 캔버스는 주로 글쓰기와 코딩을 지원하도록 설계되었으며, **편집 제안, 텍스트 길이 조정, 주석 또는 로그 추가**와 같은 특수 기능을 제공합니다.
* 🔧 캔버스 사용 시기를 인식하도록 학습된 고유한 버전의 GPT-4를 사용하면 **전체 결과물을 다시 실행하지 않고도 특정 섹션을 수정**할 수 있는 타깃 콘텐츠 업데이트가 가능합니다.
* ⚠ 캔버스의 도입은 플랫폼 리스크에 대한 과거의 경험을 떠올리게 하는 잠재적인 경쟁 위협을 의미하기 때문에 **OpenAI의 API를 기반으로 하는 AI 기업들에게 우려**가 제기되고 있습니다.
* 🎮 코딩 작업으로 Canvas를 테스트한 결과 **라이브러리 변경 및 주석 추가 등 코드를 효율적으로 수정할 수 있는 기능**이 있었지만, 때때로 의도한 것보다 더 광범위하게 변경 사항을 적용하는 경우가 있었습니다.
* 📚 창의적인 글쓰기를 위해 Canvas는 **읽기 수준을 조정**하고 **이모티콘을 추가**할 수 있지만, 이모티콘 추가의 유용성에 대한 의문이 제기되어 기능의 초기 단계 한계가 부각되고 있습니다.
* 🔍 전반적으로 캔버스는 아직 기초적인 수준이지만, 향후 발전 가능성이 높기 때문에 **AI 도구 환경에서 강력한 경쟁자**로 자리매김할 수 있으며, **플랫폼 종속성과 혁신에 대한 논의**를 촉발할 수 있습니다.

ChatGPT의 캔버스 기능은 사용자가 텍스트와 코드를 원활하게 편집하고 수정할 수 있도록 하여 협업을 혁신적으로 개선합니다.

Just yesterday, ChatGPT launched a brand new, pretty incredible feature called Canvas. This feature feels akin to ChatGPT'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 ChatGPT. You type a prompt, and instead of it just being in one single chat, it opens up this separate window where you can ask ChatGPT to edit certain parts of text or code. You can ask it to update, you can ask it to add comments, 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 ChatGPT 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 drop-down here, it is GPT-4 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. 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-4 to collaborate. So, this is not just the base GPT-4 model with some code wrapper around it; this is actually a different version of GPT-4 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-4 was trained to use Canvas. So, to do that, they developed these core behaviors: triggering the Canvas for writing and coding. 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 telepan servic, and they were truly a platform company. All they did was provide an API until they didn't, and we built an entire suest 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 ideas, 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 Open AI in the comments. All right, so we have 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 placed 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 p game 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 p game. 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 'cause 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. Thanks to the sponsor of this video, Vulture. Reap the benefits.

of the world's largest independent cloud provider. When you bring your GPU workloads over to Vulture, they have the latest Nvidia GPUs spanning 32 locations across six continents. Vulture delivers industry-leading price to performance and serious accessibility and reliability. Vulture's global fully composable cloud infrastructure moves your applications closer to your users and frees you from vendor lock-in, allowing you to bring your own networking and database solutions. And if you need to scale beyond just a single cluster, Vulture's Kubernetes engine allows you to take full control over your deployment, offering up a 100% free control plane. So if you're tired of waiting for GPUs from other providers, make sure to use Vulture. You can deploy at any scale immediately, and they have H100s, L4s, and more available right now. And you can get a fraction of a card or fully dedicated bare metal systems, which gives you full control over your hardware and your throughput. They also have one-click installation of all the applications you might need for advanced

machine learning workloads, allowing you to get up and running in minutes, not hours. So, experience the Vulture difference. Don't get bogged down by severe wait times or limited locations. Try it free today with a $300 credit for your first 30 days when you visit getvulture.com/burman and make sure to use code Burman300 at checkout to get that credit. Thanks again to Vulture for sponsoring this video. All right, now let's try fixing bugs. So, I'm actually going to switch it back 'cause 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, fixed 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. All right, 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 a 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 VSS 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 assistant. 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 looking to 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. All right, 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, etc. 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 kindergartener. 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 rotated, 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 ADD emojis 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 adore. comom

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. Contact 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. 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. Chachi PT with canvas looks very cool, still very basic, but I think 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. And if you enjoyed this video, please consider giving a like and subscribe, and I'll see you in the next one.

Huge ChatGPT Upgrade - Introducing “Canvas”

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

Matthew Berman

별도의 창에서 텍스트와 코드를 공동 편집할 수 있는 캔버스라는 새로운 기능을 ChatGPT에서 출시했습니다.

이 기능은 현재 프로 및 엔터프라이즈 계정 사용자에게 제공되며, 베타 단계 이후 무료 사용자에게도 확대할 계획입니다.

캔버스는 숙련된 버전의 GPT-4를 사용하면서 수정 사항을 제안하고, 코드를 검토하고, 목표에 맞는 업데이트를 수행하여 글 작성과 코딩을 지원하는 것을 목표로 합니다.

플랫폼 리스크의 영향과 캔버스의 기능

Twilio의 경쟁과 플랫폼 리스크는 다른 사람의 API를 기반으로 비즈니스를 구축하는 것의 위험성을 강조합니다.

0을 사용하는 캔버스는 스네이크 게임과 같은 애플리케이션을 구축하기 위한 고유한 코드 편집 환경을 제공합니다.

코드 변경, 다양한 라이브러리 사용, 실시간 업데이트 제공 등의 기능을 제공하지만 코드 실행 기능이 부족합니다.

코드 편집 및 버그 수정이 쉬워집니다.

이 도구는 강조 표시된 줄뿐만 아니라 전체 문서에 자동으로 주석을 추가합니다.

코드를 다른 언어로 포팅할 수 있는 슬라이더를 제공하여 원활하게 전환할 수 있습니다.

스폰서인 Vulture는 업계 최고의 가격 대비 성능으로 GPU 워크로드를 제공하며, 첫 30일 동안 $300 크레딧을 제공합니다.

AI의 반복적인 버그 수정 및 창의적인 글쓰기

AI는 코드의 버그를 성공적으로 수정했을 뿐만 아니라 상당한 변경도 수행하여 반복적인 개선으로 이어졌습니다.

코드에서 로그를 추가하고 제거하여 개발 프로세스를 반복할 수 있는 능력을 보여주었습니다.

창작 글쓰기에서는 AI가 지능형 딸기에 대한 스토리를 생성하고 Google 문서와 유사하게 편집 가능한 개선 제안을 제공했습니다.

나노봇으로 신경 연결 강화하기

나노봇은 원래 뇌 기능을 개선하기 위한 의학 연구용으로 설계되었으나 딸기 안에 둥지를 틀게 되었습니다.

이 로봇의 독특한 씨앗 배치는 생물체의 신경 연결을 강화하기 위한 것이었습니다.

이 도구는 길이와 읽기 수준을 조정하고 이모티콘을 추가하거나 제거하여 개인화된 느낌을 줄 수 있는 기능을 제공합니다.

발표자는 코드에서 초기화되지 않은 변수와 필드를 수정하기 위해 캔버스를 사용하고 있습니다.

이 도구는 '완료' 변수의 초기화, 연락처 정의의 널(null) 등의 문제를 성공적으로 식별하고 해결합니다.

화자는 캔버스의 버그 수정 기능이 인상적이며 다른 AI 제품과 경쟁할 수 있는 잠재력이 있다고 생각합니다.