Original video in Chinese.
Key Takeaway
- OpenAI’s newly released Canvas is an important update to ChatGPT, providing text writing and code generation features.
- Canvas borrows from Claude’s Artifacts feature and has a Grammarly-like writing style.
- The AI industry has entered the stage of commercial deployment, and white-collar workers and programmers are the main targets for AI replacement.
- Canvas’s text writing features include summarizing, expanding, adjusting style, checking text, and adding emoji.
- Canvas’s code generation features support writing code, changing programming languages, and fixing bugs.
- The article argues that the launch of Canvas poses a threat to startups, but for users who already use Artifacts and Cursor, its features are not disruptive.
There really isn’t much time left for workers.
OpenAI’s latest release, Canvas, is ChatGPT’s most important update. It brings two genuinely useful features: text writing and code generation.
When I first saw Canvas, I immediately thought of two products:
One is Claude. Canvas is essentially inspired by its Artifacts feature.
I previously mentioned in the newtype community that Artifacts is especially like a university professor’s blackboard. When you go to ask the professor a question, he pulls a clean blackboard out from behind him and explains and demonstrates everything for you in detail on it.
After Artifacts was launched, it received unanimous praise from everyone. Now OpenAI has followed suit, and it looks like this feature will become a standard for chatbot products. Maybe domestic large language model companies are already developing it.
The second is Grammarly. This is an AI writing assistant tool with 30 million daily active users.
Canvas’s text writing has a very Grammarly-like feel. If there are future updates, just copy Grammarly—it has already done deep research in this area. This track is so narrow that there isn’t much new to play with.
The reason I say there isn’t much time left for workers is because:
The entire AI industry has already entered the stage of competing on commercial deployment. Most model vendors have to rush into applications. Only a few can, like Deepseek, focus wholeheartedly on model R&D.
So which direction should they rush in to prove their phased commercial value to capital?
Pick the soft persimmon. White-collar workers are the easiest to replace and the easiest group to bully. Now that the underlying model capabilities are sufficient, as long as more effort is put into the product layer, eliminating junior white-collar workers is not a problem.
Programmers are another target. Things like code are standardized expressions, and the data volume is huge, so AI can learn them without issue. I previously introduced Cursor and Replit Agent, which are moving in this direction.
So if you can’t beat it, join it. Only workers who start fully embracing AI now and keep thinking about how to use AI will survive. You don’t need to beat AI; you only need to beat your colleagues.
Hello everyone, welcome to my channel. Humbly speaking, I’m one of the few bloggers in China who can clearly explain the why and how of AI. Remember to follow; it’s absolutely worth it. If you want to connect with me, come to the newtype community. More than 500 friends have already joined as paid members.
Back to today’s topic: ChatGPT Canvas.
This feature is still in beta and not very mature yet. For example, it is not yet a general-purpose feature; you need to switch specifically to the model version that has it. And I also ran into a situation where GPT suddenly switched to English while I was talking to it. Still, despite its flaws, it’s worth trying.
I first tested its text writing ability. I uploaded the script from the previous video and asked it to write a blog post based on that content.
At that point, an interface similar to Artifacts appeared. GPT-4o quickly finished the first draft. Notice that OpenAI has placed several customized functions in the lower right corner, including adjusting length, and so on.
When we max out the length, GPT will expand on the original basis and can increase the amount of text by up to 75%.
For the second function, at face value, you might think it means writing articles that people at different education levels can understand. But actually, it just adjusts the writing style.
For example, if you set it to kindergarten level, the article becomes especially simple and easy to understand. If you set it to the highest PhD level, it basically has an overly flowery, even a bit unnatural feel.
Honestly, OpenAI’s design here is a bit offensive. This has nothing to do with education at all. Why bother doing this?
The last two functions are straightforward. One helps you check the text to see where there are typos, where you need a subheading, and so on.
The other adds emoji to the article. A lot of people on Twitter are praising this feature. I feel nothing. I absolutely would never add emoji to any article I write.
Of course, not using these preset functions in the lower right is also fine. We can select any part of the content and directly give GPT a request. For example, make the opening more engaging. Or elevate the ending a bit more.
That’s all for Canvas’s text writing ability: simple and basic. This kind of capability can be achieved through prompting and Agent Workflow. OpenAI has packaged them all into the product and even added a little gimmick like emoji insertion to please users. You can see they’re still very slick.
The code generation side works in pretty much the same way.
As a test, I asked GPT to help me write a Snake mini-game. Again, an expandable button appeared in the lower right corner. The option to change programming language is kind of interesting. For example, we can have GPT convert the code it has already generated into JavaScript, TypeScript, and several other mainstream languages.
The remaining functions are very simple, such as fixing bugs. I won’t go into detail; you can understand them at a glance.
That’s all there is to Canvas for now. I’m guessing some startups are definitely not going to sleep well. OpenAI is exchanging their lives for time for itself. As a user, I think it’s fine. Maybe that’s because I’ve been using Artifacts and Cursor all along, so I’m not that excited seeing this. In order to try Canvas, I bought a one-month membership and don’t plan to renew. I’ll keep using Claude after this.
OK, that’s it for this video. If you want to discuss AI, come to the newtype community. Then I’ll see you in the next one!