Training would drop timestamps half the time for robustness.
i get the feeling messages in chat/codex don’t have timestamps still, model doesn’t have a good sense for time? bump to oai friends: if so i have a renderer pr from a while back, codex could 1 shot. just edit the renderer, drop it out half the time, run a derisk would be so nice!
@willdepue great minds
reasonably i understand people are hesitant about splitbraining — old data might not have timestamp info & i don't think the api would want to support this — but i'm pretty sure splitbraining already happens/it'll be ok if you just drop it out aggressively!
i get the feeling messages in chat/codex don’t have timestamps still, model doesn’t have a good sense for time? bump to oai friends: if so i have a renderer pr from a while back, codex could 1 shot. just edit the renderer, drop it out half the time, run a derisk would be so nice!
Training would drop timestamps half the time for robustness.
i get the feeling messages in chat/codex don’t have timestamps still, model doesn’t have a good sense for time? bump to oai friends: if so i have a renderer pr from a while back, codex could 1 shot. just edit the renderer, drop it out half the time, run a derisk would be so nice!
Users react to OpenAI's missing timestamps in Chat and Codex messages, with some endorsing the flag as insightful while others dismiss the issue as typical ongoing shortcomings.
@willdepue great minds
reasonably i understand people are hesitant about splitbraining — old data might not have timestamp info & i don't think the api would want to support this — but i'm pretty sure splitbraining already happens/it'll be ok if you just drop it out aggressively!
i get the feeling messages in chat/codex don’t have timestamps still, model doesn’t have a good sense for time? bump to oai friends: if so i have a renderer pr from a while back, codex could 1 shot. just edit the renderer, drop it out half the time, run a derisk would be so nice!