news

there is no disruptive innovation, and the new features of chatgpt canvas, which is rushing towards the productivity revolution, vaguely reveal the shadow of claude artifacts.

canvas is a new interface designed to help users collaborate more closely with chatgpt on writing and coding.

regarding this feature, openai research director karina nguyen explained her thoughts on the x platform:

the ultimate agi interface in my mind is a blank canvas.

it can evolve and transform itself in line with human preferences, and invent new ways of interacting with humans, redefining our relationship with ai technology and the internet as a whole.

in short,Canvas it is a larger collaboration space that is more suitable for complex work.

interestingly,on the openai official website, we also found that the leadership team supporting this new feature includes some familiar faces who left this year-barret zoph, john schulman, mira murati.

starting today, openai will launch canvas to chatgpt plus and team users.

enterprise and edu users will gain access next week.after testing, openai plans to roll out canvas to free users.

i changed what i pointed out, and my writing ability became stronger.

writing and coding are two high-frequency usage scenarios of chatgpt.

although the chat interface is simple and efficient and suitable for a variety of tasks, it currently seems to be somewhat inadequate when handling projects that require complex steps and multi-faceted skills.

canvas provides a new work interface where users can edit and improve the output of ai.

openai officials said that canvas can make chatgpt understand your thoughts better. for example, you can highlight key parts and tell chatgpt where to pay special attention.

in official terms, it's like having a copy editor or code reviewer next to you providing real-time feedback and suggestions in real time.

in the canvas interface, you can also easily edit text or code.

the page also comes with shortcuts that let you ask chatgpt to adjust the length of your writing, debug your code, and quickly perform other useful actions. undo with one click and restore to the original version is a breeze.

writing shortcuts include:

there is no need to kill a chicken with a big knife. frequently calling canvas for small things is a bit complicated.

normally, canvas is automatically enabled when chatgpt detects a scenario where it can help. of course, you can also add "use canvas" when asking a question to turn on this feature.

according to openai, canvas will automatically start when encountering a request like "write a blog post about the history of coffee beans."

but if it's a simple question-and-answer task, such as "help me make a new dinner recipe," launching canvas is a bit fussy.

why can the canvas function improve writing quality and interactive experience?

officials said that similar to a quality monitoring system, the development team set up more than 20 automated tests to track the performance of their ai during the development process.

during the development process, the research team used the recently launched openai o1-preview version to generate synthetic data, which was then used as a post-training data set to enhance key functions of the model, such as writing and interaction with new users.

real ai programmers are here

consider that coding is usually not done in one go, but requires multiple revisions and improvements.

the purpose of canvas is to make it easier and clearer to track and understand the modifications made by chatgpt, and in the future openai will further consider improving the transparency of the editing process.

coding shortcuts include:

in order to avoid disturbing experienced developers during coding tasks, canvas should not be called so frequently. openai has specifically made improvements in this area, allowing it to trigger canvas more accurately during writing and coding tasks, achieving 83% and 94% accuracy respectively.

currently, canvas built based on the gpt-4o model is still in the early testing stage. users can switch to experience this feature in the model menu bar. openai claims that it will quickly improve its functionality.

in addition to teaching the model to boot properly, the research team also taught it how to edit content smarter.

when canvas is started, the model needs to decide whether to modify only certain parts of the content or rewrite all of it.

therefore, if the user explicitly points out what needs to be changed on the interface, the model will only modify those parts. and if the user doesn't specify it, the model will choose to rewrite the entire content.

the ultimate agi interface - the blank canvas

Karina Nguyen some practical use cases of canvas were also shared.

gpt-4o can be combined with canvas to research art history and write reports. you can ask to verify your opinions and add citations.

gpt-4o combined with canvas can write rust code and conduct code review.

after browsing recipes, canvas can create new recipes based on existing ingredients.

the carnival at the openai developer conference continues.

the focus of canvas functionality is still firmly locked in the content creation and coding development process.

the design freedom of independent windows is higher, allowing us to interact and manage complex content directly on the main interface. this not only greatly optimizes the user experience, but also allows us to see the future of human-computer interaction.

ai services with only simple question and answer functions can no longer keep up with the first echelon of human “co-pilots”.

with the popularity of functions/tools such as claude artifacts and cursor, the launch of the canvas function is also an important part of the market competition.

openai's "ammunition arsenal" for human-computer interaction is still sufficient, and it is still very successful.