LLMs Improve Planning with Self-Critique
Analysis
This paper demonstrates a novel approach for improving Large Language Models (LLMs) in planning tasks. It focuses on intrinsic self-critique, meaning the LLM critiques its own answers without relying on external verifiers. The research shows significant performance gains on planning benchmarks like Blocksworld, Logistics, and Mini-grid, exceeding strong baselines. The method's focus on intrinsic self-improvement is a key contribution, suggesting applicability across different LLM versions and potentially leading to further advancements with more complex search techniques and more capable models.
Key Takeaways
- •LLMs can improve planning performance through intrinsic self-critique.
- •The method achieves state-of-the-art results on considered models.
- •The approach is applicable across different LLM versions.
- •Iterative correction and refinement further enhance performance.
Reference / Citation
View Original"The paper demonstrates significant performance gains on planning datasets in the Blocksworld domain through intrinsic self-critique, without external source such as a verifier."