4 Comments
User's avatar
avi's avatar

Nice. Does this work only with reasoning models ? Or any LLM? And have you experimented with fine tuning an LLM to implicitly do this cot ?

Expand full comment
Zachary Huang's avatar

It works for any LLMs.

Training an LLM to do *GREAT REASONING** is very promising (e.g., deepseek R1) but also complex and costly.

Expand full comment
Alex Rosenfeld's avatar

Really appreciate your ability demonstrate the effectiveness of this flow with such an elegant and simple framework!

Expand full comment
Zachary Huang's avatar

Thank you!

Expand full comment