註冊並分享邀請連結,可獲得影片播放與邀請獎勵。

Bryan Bischof fka Dr. Donut
@BEBischof
Leading AI @theoryvc; Teach ML @rutgersu. Prev: Led AI @_hex_tech; Led Data @weights_biases, ML @stitchfix, Led Data @bluebottleroast
加入 December 2010
2.8K 正在關注    5.4K 粉絲
I know this is about thinking traces but before thinking traces this was a really effective tool for improving cot in tool calling cases. After thinking models this trick became useful for explaining caveats or uncertainties.
顯示更多
guys you do know you can just disable thinking, and instead give it a "deep_think" tool, and it will call it with internal CoT reasoning format right? gl fixing that