&&
@sithis3.bsky.social
Reposted by &&
This post from Jan Leike is a decent summary of where post training is at. Easy to solve tasks with strict verifiers, but fuzzy tasks are hard.
aligned.substack.com/p/crisp-and-...
aligned.substack.com/p/crisp-and-...
Crisp and fuzzy tasks
Why fuzzy tasks matter and how to align models on them
aligned.substack.com
November 23, 2024 at 4:38 PM
This post from Jan Leike is a decent summary of where post training is at. Easy to solve tasks with strict verifiers, but fuzzy tasks are hard.
aligned.substack.com/p/crisp-and-...
aligned.substack.com/p/crisp-and-...