@Dorialexander This is a great point. I suspect there isn’t too much of a delta on this between Chinese and US labs tho (my guess is Anth leads here by a bit).
Actual leapfrog: they are simulated (reminder from GLM 5)
Users praise the observation on Chinese labs using simulated terminal environments for RL training as a great point with little gap to leading US labs.
Based on 1 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@Dorialexander This is a great point. I suspect there isn’t too much of a delta on this between Chinese and US labs tho (my guess is Anth leads here by a bit).
Actual leapfrog: they are simulated (reminder from GLM 5)
Actual leapfrog: they are simulated (reminder from GLM 5)
Does anyone know what the data landscape is like at the Chinese labs? Are they buying as many RL environments as OAI/Anth, and is it just cheaper to get that in China? Or do they compensate via distillation?
Users praise the observation on Chinese labs using simulated terminal environments for RL training as a great point with little gap to leading US labs.
Based on 1 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.
