News

  • News
  • Critic-CoT Framework Enhances Reasoning in Large Language Models
Share

Critic-CoT Framework Enhances Reasoning in Large Language Models

21h ago
Chains of Thought Critic-CoT Large Language Models Reasoning Self-Critique Iterative Refinement GSM8K MATH
The Critic-CoT framework improves the reasoning capabilities of large language models through self-critique and iterative refinement, significantly boosting task-solving performance on datasets like GSM8K and MATH.

Critic-CoT Framework Enhances Reasoning in Large Language Models

Chains of Thought (CoT) reasoning models, such as the Critic-CoT framework, are designed to enhance the reasoning abilities of large language models (LLMs) through self-critique and refinement. The Critic-CoT approach pushes LLMs toward System-2-like critic capabilities, enabling them to engage in slow, analytic self-critique and iterative refinement. This is achieved through a step-wise CoT reasoning paradigm and the automatic construction of distant-supervision data without human annotation.

Experiments on datasets like GSM8K and MATH have demonstrated that Critic-CoT significantly boosts task-solving performance by filtering out invalid solutions and refining reasoning processes. The framework also investigates the intrinsic correlation between critique and task-solving abilities within LLMs, revealing that these abilities can mutually reinforce each other rather than conflict.

For more detailed information, you can refer to the arXiv paper and the OpenReview submission.

Sources

Critic-CoT: Boosting the reasoning abilities of large language model ... Self-critic has become a crucial mechanism for enhancing the reasoning performance of LLMs.
Critic-CoT: Boosting the reasoning abilities of large language model ... This paper proposes Critic-CoT, a method designed to improve LLMs' reasoning by developing a stepwise, chain-of-thought (CoT) critique process.
Detecting misbehavior in frontier reasoning models - OpenAI Chain-of-thought (CoT) reasoning models “think” in natural language understandable by humans. Monitoring their “thinking” has allowed us to ...
2025© BestAI