Latent Veracity Inference for Identifying Errors in Stepwise Reasoning
This addresses the issue of reduced performance and trustworthiness in language models for users relying on stepwise reasoning, though it is incremental as it builds on existing CoT methods.
The paper tackles the problem of inaccurate statements in Chain-of-Thought reasoning by proposing latent veracity variables and a search algorithm (Veracity Search) for inference-time verification, achieving reliable error identification across logical, mathematical, and commonsense benchmarks with comparable zero-shot accuracy.
Chain-of-Thought (CoT) reasoning has advanced the capabilities and transparency of language models (LMs); however, reasoning chains can contain inaccurate statements that reduce performance and trustworthiness. To address this, we propose to augment each reasoning step in a CoT with a latent veracity (or correctness) variable. To efficiently explore this expanded space, we introduce Veracity Search (VS), a discrete search algorithm over veracity assignments. It performs otherwise intractable inference in the posterior distribution over latent veracity values by leveraging the LM's joint likelihood over veracity and the final answer as a proxy reward. This efficient inference-time verification method facilitates supervised fine-tuning of an Amortized Veracity Inference (AVI) machine by providing pseudo-labels for veracity. AVI generalizes VS, enabling accurate zero-shot veracity inference in novel contexts. Empirical results demonstrate that VS reliably identifies errors in logical (ProntoQA), mathematical (GSM8K), and commonsense (CommonsenseQA) reasoning benchmarks, with AVI achieving comparable zero-shot accuracy. Finally, we demonstrate the utility of latent veracity inference for providing feedback during self-correction and self-improvement.