news

focus on

openai's latest new interface, canvas, aims to transform the function of chatgpt from a simple chat interaction into a more collaborative work space, especially suitable for writing and coding.

different from the regular chat window, canvas opens up an independent workspace and is equipped with a chat window on the right side, allowing users to work together with chatgpt to refine and edit ideas directly in documents or code. with this innovative feature, openai effectively solves the core problems of previous interfaces: frequent chat round-trips, redundant and repetitive output, and character response limitations. these problems are not conducive to long-term processing of complex and multi-step projects.

canvas has been available to plus users since thursday, local time in the united states, although the demo clearly demonstrated its power and limitations. this new user experience is undoubtedly a response to other companies’ existing practices in the field of artificial intelligence-enhanced productivity. long before canvas came out, google and microsoft were already exploring integrated applications of artificial intelligence, as evidenced by gemini in google docs and copilot in microsoft word. in fact, both gemini and copilot adopt nearly identical strategies in document editing, aiming to promote collaboration between humans and artificial intelligence.

what exactly is canvas?

canvas introduces a new user experience and workflow to chatgpt that is unique in its ability to understand and adapt to the specific content the user is trying to accomplish. it works like a collaborative editor, providing instant feedback directly on text or code. openai points out that for writers, canvas can suggest editorial adjustments, optimize document length, and even adjust tone based on reading level to better suit different audiences.

for programmers, the tool can inspect code, add debug logs and comments, and translate code between programming languages ​​when necessary. openai claims that "canvas can provide inline feedback and suggestions for the entire project," thereby greatly enhancing the interactive experience between users and artificial intelligence in a way that traditional chat interfaces cannot achieve.

rather than iterating over chatgpt's output or asking it to rewrite specific sections in a prompt, canvas allows users to select text directly from a live document and ask ai to help improve the selection. this feature makes the canvas experience more similar to mainstream office software such as google docs and word.

according to openai, canvas has been available to chatgpt plus and team users starting from thursday local time, while enterprise and education users will gain access next week. a wider rollout plan for all chatgpt free users will be implemented after the testing phase is completed. currently, canvas is an exclusive perk for gpt-4 users, and chatgpt automatically launches canvas when it detects a writing or coding scene. in addition, users can also manually activate this feature by entering "use canvas" in the prompt box.

copying google and microsoft?

it is not difficult to find from the demonstration that when users are committed to writing long documents or codes, canvas shows greater advantages compared to the current mode. however, in terms of the breadth of writing and the diversity of word processing, canvas' simplicity and elegance make it slightly more limited than gemini in google docs. the sidebar of google docs is similar to canvas in layout, but the former provides richer options, such as adjusting tone, summarizing key points, creating bulleted lists, elaborating, shortening content, rephrasing, and an open style prompts that can respond to various user needs, such as converting prose into poetry.

although google docs may be more powerful in terms of functionality, its interface design is also more complex and slightly confusing than canvas. google docs, for example, uses pop-ups to display the results of commands on a user-selected portion of text, giving users the opportunity to preview and adjust the command before applying it to the page. this modal window gives the user the option to further refine the output before replacing the selected text or inserting it as a new paragraph.

in contrast, the features of chatgpt canvas are particularly intuitive. it allows users to edit text directly, as if there is an invisible collaborative editor silently undertaking the writing task. this experience makes the final result appear more precise and engaging.

for writers, the simplicity and neatness of the interface will undoubtedly help them quickly get into a creative state, focus their attention, and avoid the interference of redundant user interface elements in traditional word processing programs. take ia writer, a minimalist word processing software specially built for writers as an example. it almost restores the pure experience of writing on a typewriter, while incorporating the convenience brought by digitalization. canvas is similar to ia writer to some extent, but the difference is that it also incorporates the assistance of artificial intelligence.

when canvas recognizes a coding task, it can seamlessly switch to "programming mode" to provide users with code inspection, debugging, and cross-language portability assistance. openai seemed to hint in its blog post that the company will continue to embrace and adapt to this shift in personalization, making flexible adjustments based on the actual needs of users. perhaps in the near future, canvas can automatically transform from "novel creation mode" to "academic paper mode" and intelligently adjust the available tools in the sidebar. this is undoubtedly a potential and unique user experience development path.

for now, canvas seems to be a strong response to the limitations of chatgpt’s current user experience. it creates a more focused and refreshing work environment, but its core ideas draw heavily on the existing practices of google and microsoft, rather than creating a truly disruptive user experience to solve similar problems. although canvas is a pretty cool feature that will be very attractive to loyal users of chatgpt, it remains to be seen whether it will be enough to fully replace traditional word processors or programming environments with the power of artificial intelligence.

benchmarking anthropic artifacts?

canvas' separate window and model collaboration capabilities are reminiscent of another chatbot, anthropic's claude artifacts, which also has a separate window that clearly displays any changes made through new prompts.

claude artifacts regards easy-to-access and intuitively understood interfaces as the core functions of chatbots, and declares this to be "the most important artificial intelligence development this year." unlike canvas, artifacts is already available to all claude chat users.

in addition, artifacts also allows users to intuitively see the effect of the code they generate. users can instruct claude to write code, edit code, and instantly view the results, such as a prototype of a website or game. in contrast, canvas mainly shows the user the generated code and the editing around it.

the new interface battleground, led by openai’s canvas and anthropic’s artifacts, does shed light on a problem that third-party ai applications have long grappled with: how to make chatbots easier to read and use.

applications such as hyperwrite, jasper, and jotbot all have text generation and editing capabilities. many other software incorporate similar code and text editing features, with the core idea being that users can modify their work without leaving the chat window.

eventually, features like canvas and even artifacts may become commonplace as people increasingly expect a smoother, more collaborative work experience with chatbots. (tencent technology specially compiled by jinlu)