Answering the Unanswerable Is to Err Knowingly: Analyzing and Mitigating Abstention Failures in Large Reasoning Models
This addresses a critical issue for trustworthy AI by mitigating abstention failures in LRMs, though it is incremental as it builds on existing cognitive monitoring and intervention techniques.
The paper tackles the problem of large reasoning models (LRMs) failing to abstain from answering inherently unanswerable questions, such as math problems with insufficient conditions, and proposes a lightweight two-stage method that significantly improves the abstention rate while maintaining overall reasoning performance.
Large reasoning models (LRMs) have shown remarkable progress on complex reasoning tasks. However, some questions posed to LRMs are inherently unanswerable, such as math problems lacking sufficient conditions. We find that LRMs continually fail to provide appropriate abstentions when confronted with these unanswerable questions. In this paper, we systematically analyze, investigate, and resolve this issue for trustworthy AI. We first conduct a detailed analysis of the distinct response behaviors of LRMs when facing unanswerable questions. Then, we show that LRMs possess sufficient cognitive capabilities to recognize the flaws in these questions. However, they fail to exhibit appropriate abstention behavior, revealing a misalignment between their internal cognition and external response. Finally, to resolve this issue, we propose a lightweight, two-stage method that combines cognitive monitoring with inference-time intervention. Experimental results demonstrate that our method significantly improves the abstention rate while maintaining the overall reasoning performance.