It's been five days since OpenAI's developer day, and Tibo has announced the next generation of models.
This time around, they're finally going to write the hill code for the pipeline model.
Recently, Tibo Sottiaux, Head of Product at OpenAI, posted that future models will be better at removing and simplifying code.
In the end, there is a long saying: "It's about time".
This does not seem to be a random prediction.
OpenAI researcher Rajan Agarwal then re-posted the post, saying, "This is what I'm doing day and night, and I'm looking forward to seeing if you can feel the difference".
Rajan is currently in charge of coding post-training, focusing on making coding models and Codex capable of writing code that is more "qualified" than coding capabilities.
A few hours later, he flipped out the same Tibo post and asked the user directly: "Did you ever see sloppy code written on our model? What's your biggest problem with code quality?"
Is the Astra 6.1 coming?
Tibo：6.1 coming soon
Recently, the Wall Street Journal reported that OpenAI had originally planned to launch the Astra 6.1 in October, but was temporarily suspended after internal security testing.
In testing, this version has shown to be uncomfortable at the "limit of authority", sometimes continuing to perform operations without user permission, and not being candid enough about exactly what they did.
OpenAI ultimately decided not to publish it first and continued to address these security and coordination issues.
But just now, in a post praising the Astra 6, Tibo left a vague comment:
「6.1 coming soon」
Combined with the code removal capability that was announced hours ago, it seems that the real level of the new generation of models is slowly surfacing.
The codex is the next step, the netizens say?
It's unclear how the new model will work, but it looks like the product update for OpenAI is "The Crowd Route".
Similarly, today, Tibo published two articles seeking advice from netizens about updates to the codex and sidebar.
The first thing the commenters said was to add the reset button (hahahaha).
It is hoped that the Codex will truly support the transfer of tasks between Agents: one agent is responsible for development, another is responsible for review, then handed over to other Agents, repaired, tested, and finally delivered, without the need for manual transportation of the context.
Others want to directly open the memories of Codex and ChatGPT. ChatGPT already knows its own projects, habits and context, but it has to start over again.
There are many similar needs: unified Chat, Work and Codex interfaces; adding large amounts of chat to folder levels and tag pages; allowing the Codex Agent to collaborate directly on different computers; and even allowing users to plug in, interrupt, and then let the Agent continue to work for half the time.
It has also been suggested that the Codex, like a "project historian", should automatically remember where a long-term project has progressed and what has not been completed.
When it comes to the sidebar, someone who is most concerned about usability wants the sidebar to be directly integrated into Git management so that you don't have to open the IDE separately to see the version status of the branch.
Some people also want to make it a project dashboard: put a token with an API, track costs, feature lists, task time, and see where a project is going at any time.
Others want the new sidebar to take on the role of organizing and switching, such as breaking chats into different categories, switching with icons, and even integrating browsers to avoid having to reboot a chat to open a web page.
So the question arises: if you could only choose one, which would you most want to update first, the Codex or the next generation of models?
The link is as follows: