AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions

Polina Kirichenko (New York University) · Mark Ibrahim (Fundamental AI Research (FAIR) at Meta) · Kamalika Chaudhuri (FAIR, Meta and UCSD) · Samuel J. Bell (Meta FAIR)
abstentioncomplex problem solvingevaluation frameworkfalse premisesill-posed questionsllm reliabilitymodel scalingoutdated informationreasoning fine-tuningsubjective interpretationssystem promptsystematic evaluationuncertainty reasoningunderspecified queries

For Large Language Models (LLMs) to be reliably deployed in both everyday and high-stakes domains, knowing when not to answer is equally critical as answering correctly.Real-world user queries, which can be underspecified, ill-posed, or fundamentally unanswerable, require LLMs to reason about uncertainty and selectively abstain