NEW Research from Meta Superintelligence Labs and collaborators. The default approach to improving LLM reasoning today remains extending chain-of-thought sequences. Longer reasoning traces aren't always better. Longer traces conflate reasoning depth with sequence length and
Meta Research Challenges Chain-of-Thought Reasoning Approach
By
–
