news

author of this article: li dan

source: hard ai

in addition to the general dialogue window, openai also provides users of chatgpt with a new window that is more conducive to writing and coding project work.

on thursday, october 3rd, eastern time, openai announced the launch of a new interface called canvas, which is built with openai's latest main model gpt-4o and is used to use chatgpt to collaborate on writing and coding project work that is not limited to chatting. canvas opens in a separate window, allowing users chatgpt to collaborate on a project.

some media commented that several consumer ai vendors have launched editable workspaces as a practical way to use generative ai.OpenAIit is working hard to copy its competitors' products and launch its own new features in chatgpt to expand its paying user base. canvas is similar to artifacts, a dynamic workspace function developed by openai's rival anthropic outside of the chat interface, and also a bit like the popular free code generation tool cursor.

write with editing suggestions, adjust length, choose reading level, add final touches and emoticons

openai said that canvas introduces a new way of collaboration - not just through dialogue, but through users and ai working side by side to create and improve ideas. with the help of canvas, chatgpt can better understand the context of the task to be completed by the user. users can highlight specific sections to indicate exactly what they want chatgpt to focus on. just like a copy editor or code reviewer, it can provide inline feedback and suggestions with the entire project in mind.

in terms of writing, some shortcuts provided by canvas include:

• suggested editors: chatgpt provides inline suggestions and feedback.
• adjust length: edit the document length, shorten or lengthen it.
• change the reading level: adjust the reading level to meet the needs of people with multiple reading abilities from kindergarten to postgraduate level.
• add final touches: check for grammar, clarity, and consistency.
• add emoticons: add relevant emojis for emphasis and color.

openai product manager daniel levine demonstrated to techcrunch how to use canvas to help write emails. users can prompt chatgpt to generate an email, which then pops up in the canvas window. users can toggle sliders to adjust the length of their writing, highlight specific sentences, and ask chatgpt to make changes, such as asking it to sound friendlier or adding emoticons. users can also ask chatgpt to rewrite the entire email intact in another language.

help review code, add logs and comments, modify errors, convert to python and other languages

openai introduced how to write code in the canvas interface, saying that coding is an iterative process and tracking all revisions of the code in chat can be difficult, while canvas makes it easier for users to track and understand changes to chatgpt. openai plans to continue to increase the transparency of such edits.

encoding shortcuts include:

• review code: chatgpt provides inline suggestions for improving users' code.
• add logging: insert statements to help users debug and understand the code.
• add comments: add comments to the code to make it easier to understand.
• fix errors: detect and rewrite problematic code to resolve errors.
• port to language: translate user's code into javascript, typescript, python, java, c++ or php.

for coding, levine demonstrated asking chatgpt to create an api web server in python. the server code is generated in the canvas window. press the "add comment" button and chatgpt will add inline documentation explaining the code in plain english. users can highlight a piece of code created by chatgpt and have it explain it or ask questions about it. chatgpt also has a "review code" button, which will suggest specific edits to the code in the window, whether generated by gpt or written by the user themselves, and allow the user to approve, edit themselves, or reject the suggestion. if the user presses approve, chatgpt will try to fix the error itself.

some netizens summarized some features of canvas:

users can edit canvas directly or ask ai to edit (it feels like openai is following google docs)
edits are targeted at specific parts rather than full regeneration (faster and with fewer errors)
can undo changes (a good alternative to repeated prompts)
provides some tool suggestions such as "adjust length". it would be really cool when these suggestions are generated based on text

ethan mollick, a professor at wharton school who studies ai, innovation, and startups, commented that openai’s canvas does a good job with the interface, and everything is real-time and smooth. but it brings home the fact that we don’t yet know what the right approach is to co-writing with ai. that being said, canvas makes work feel more integrated and robot-like, rather than a human-machine takeover.