Dancing with Critiques: Enhancing LLM Reasoning with Stepwise Natural Language Self-Critique
This addresses the problem of nuanced feedback in LLM reasoning for researchers and practitioners, offering a broadly applicable method without task-specific training, though it is incremental as it builds on existing inference-time scaling approaches.
The paper tackles the challenge of enhancing large language models' reasoning for complex multi-step tasks by introducing stepwise natural language self-critique (PANEL), which uses self-generated critiques instead of scalar rewards to guide inference, resulting in significant performance improvements on benchmarks like AIME and GPQA.
Enhancing the reasoning capabilities of large language models (LLMs), particularly for complex tasks requiring multi-step logical deductions, remains a significant challenge. Traditional inference time scaling methods utilize scalar reward signals from process reward models to evaluate candidate reasoning steps, but these scalar rewards lack the nuanced qualitative information essential for understanding and justifying each step. In this paper, we propose a novel inference-time scaling approach -- stepwise natural language self-critique (PANEL), which employs self-generated natural language critiques as feedback to guide the step-level search process. By generating rich, human-readable critiques for each candidate reasoning step, PANEL retains essential qualitative information, facilitating better-informed decision-making during inference. This approach bypasses the need for task-specific verifiers and the associated training overhead, making it broadly applicable across diverse tasks. Experimental results on challenging reasoning benchmarks, including AIME and GPQA, demonstrate that PANEL significantly enhances reasoning performance, outperforming traditional scalar reward-based methods. Our code is available at https://github.com/puddingyeah/PANEL to support and encourage future research in this promising field.