Tibo Sottiaux, OpenAI Codex engineering lead, posts master plan for releasing better models, shipping weekly product updates, and securing additional compute
kache replies that the plan is the best available but improvable
Many users praised the plan for frequent AI model and product releases as a recipe for real progress, while others objected to rising costs, lost usage limits, and recent model changes.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@thsottiaux I'm finding that 5.5 is not very good at automated RL research. Specifically puffer lib. One of the failure cases is that it gets confused with total reward going up when reviewing experiments which have differently tuned reward scales
Our master plan is to release better and more efficient models. And also to release better products, week after week. Oh and get more compute too. Together with spending too much time on x. How good is this plan?
@thsottiaux OpenAI employee walking toward the quota reset button like:
“should we…?”
everyone on X screaming YES in 47 languages 😭
@thsottiaux it’s a good plan sir it’s an ok tweet
@thsottiaux plans good, but the following is even better tbh
holding you to it bro
@thsottiaux Sounds good
@thsottiaux When I say not very good I mean the best. But could be better
@thsottiaux I'm finding that 5.5 is not very good at automated RL research. Specifically puffer lib. One of the failure cases is that it gets confused with total reward going up when reviewing experiments which have differently tuned reward scales
@thsottiaux I can work around it with clever harnessing and prompting.
@thsottiaux When I say not very good I mean the best. But could be better
@thsottiaux Yes along with separate changelogs for codex on Mac vs windows!
@thsottiaux sounds great, so when is the new model coming?
@thsottiaux This plan can be improved by doing about two resets per week.
I’d say it’s an amazing plan. At this point, the only thing keeping me tied to Google is a really good video understanding model.
No other company has added video as part of their multi-modal model except Google.
So it would be awesome if you guys got that in there at some point.
@thsottiaux sick of the daily updates. get a life
@thsottiaux Wouldn't it be better to stop posting ambiguous information and only post actual decisions?
@thsottiaux indeed a good plan, but coders now care much about limit rate and saving tokens
@mwfowlie @thsottiaux they hinted at that, I think its coming tomorrow or the day after
plus they have yet to celebrate 5m?
@thsottiaux If we are doing Codex Thursdays, what are the odds we get ChatGPT Tuesdays?
@thsottiaux @Grok Just take Meta’s GPUs.
@thsottiaux When is realtime becoming a reality ? 😉
Better + more efficient models + weekly product drops is exactly what devs need. The weekly shipping vibe is refreshing.
Quick wishlist from a heavy Codex user in India:
• More generous rate limits / weekly resets (or region-aware pricing) — we’re grinding hard here • Faster rollout of Realtime Voice + agent features • Keep crushing the “Codex anywhere” vision Keep cooking.
This plan is working.
@thsottiaux This plan doesn't seem to include distributing tokens to the poor, ending world token hunger once and for all - Tibo got hacked.