undefined | Better HN

0 pointsregularfry2mo ago0 comments

Thinking vs non-thinking. There'll be a token cost there. But still fairly remarkable!

0 comments

1 comments · 1 top-level

Is there a reason we can't use thinking completions to train non-thinking? i.e. gradient descent towards what thinking would have answered?

j / k navigate · click thread line to collapse