Probabilistic Soundness Guarantees in LLM Reasoning Chains
This addresses reliability issues in LLM reasoning for users relying on automated reasoning systems, though it is an incremental improvement over existing error detection methods.
The paper tackles the problem of error propagation in LLM reasoning chains by introducing ARES, a probabilistic framework that provides certified soundness guarantees, achieving state-of-the-art performance with 72.1% Macro-F1 and up to 90.3% F1 on specific tasks.
In reasoning chains generated by large language models (LLMs), initial errors often propagate and undermine the reliability of the final conclusion. Current LLM-based error detection methods often fail to detect propagated errors because earlier errors can corrupt judgments of downstream reasoning. To better detect such errors, we introduce Autoregressive Reasoning Entailment Stability (ARES), a probabilistic framework that evaluates each reasoning step based solely on previously-verified premises. This inductive method yields a nuanced score for each step and provides certified statistical guarantees of its soundness, rather than a brittle binary label. ARES achieves state-of-the-art performance across four benchmarks (72.1% Macro-F1, +8.2 points) and demonstrates superior robustness on very long synthetic reasoning chains, where it excels at detecting propagated errors (90.3% F1, +27.6 points).