Reactions from ranked influencers
2 posts
Kimi team: this kind of thing is a goldmine for GRM training Sol (as well as Fable) look at K3's reasoning chain and notices an error
@teortaxesTex There is literally no way to stop this huh
Reactions from the X community
@teortaxesTex I've noticed it many times in coding agents. The CoT starts having completely wrong assumptions, to the point it really gets on my nerves, but steering/correcting it often doesn't help. If you let the model continue, instead, it may reach a reasonable outcome.
@teortaxesTex This kind of self-correction is valuable for improving future AI reasoning models.
View all
Kimi team: this kind of thing is a goldmine for GRM training Sol (as well as Fable) look at K3's reasoning chain and notices an error
@teortaxesTex There is literally no way to stop this huh