Balancing Faithfulness and Performance in Reasoning via Multi-Listener Soft Execution

TL;DR

This paper introduces REMUL, a multi-listener reinforcement learning method that enhances faithfulness and interpretability of reasoning traces in large language models without sacrificing task performance.

Contribution

It proposes a novel multi-party RL approach that improves reasoning trace faithfulness and interpretability while maintaining or enhancing accuracy in reasoning tasks.

Findings

REMUL improves faithfulness metrics across multiple benchmarks.

It enhances reasoning trace clarity and legibility.

It maintains or improves task accuracy.

Abstract

Chain-of-thought (CoT) reasoning sometimes fails to faithfully reflect the true computation of a large language model (LLM), hampering its utility in explaining how LLMs arrive at their answers. Moreover, optimizing for faithfulness and interpretability in reasoning often degrades task performance. To address this tradeoff and improve CoT faithfulness, we propose Reasoning Execution by Multiple Listeners (REMUL), a multi-party reinforcement learning approach. REMUL builds on the hypothesis that reasoning traces which other parties can follow will be more faithful. A speaker model generates a reasoning trace, which is truncated and passed to a pool of listener models who "execute" the trace, continuing the trace to an answer. Speakers are rewarded for producing reasoning that is clear to listeners, with additional correctness regularization via masked supervised finetuning to counter the tradeoff between faithfulness and performance. On multiple reasoning benchmarks (BIG-Bench Extra Hard, MuSR, ZebraLogicBench, and FOLIO), REMUL consistently and substantially improves three measures of faithfulness -- hint attribution, early answering area over the curve (AOC), and mistake injection AOC -- while also improving accuracy. Our analysis finds that these gains are robust across training domains, translate to legibility gains, and are associated with shorter and more direct CoTs.