Chain-of-Thought

SEAL: Steerable Reasoning Calibration of Large Language Models for Free

A training-free approach that calibrates chain-of-thought reasoning in LLMs, improving accuracy while reducing computational overhead.