AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions

2025-06-10

AI Summary

Current large language models (LLMs) lack reliable abstention capability—i.e., the ability to deliberately withhold answers to unanswerable questions (e.g., those with false premises, unknown answers, or semantic ambiguity)—and no systematic evaluation framework exists. Method: We introduce AbstentionBench, the first large-scale benchmark for abstention evaluation, covering five categories of unanswerable questions and assessing 20 state-of-the-art LLMs across 20 heterogeneous datasets. Contribution/Results: Key findings reveal that reasoning-oriented fine-tuning degrades abstention performance by 24% on average, exposing a fundamental deficiency in LLMs’ uncertainty reasoning. Neither scaling model size nor applying sophisticated system prompts yields significant improvement. Our work establishes the first multi-dimensional, empirically grounded abstention evaluation paradigm—providing a foundational benchmark, actionable insights, and methodological scaffolding for developing trustworthy LLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty
• Natural Language Processing: (Large) Language Models

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

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 -- i.e., refuse to answer definitively. However, abstention remains understudied, without a systematic evaluation framework for modern LLMs. In this work, we introduce AbstentionBench, a large-scale benchmark for holistically evaluating abstention across 20 diverse datasets, including questions with unknown answers, underspecification, false premises, subjective interpretations, and outdated information. Evaluating 20 frontier LLMs reveals abstention is an unsolved problem, and one where scaling models is of little use. While recent reasoning LLMs have shown impressive results in complex problem solving, surprisingly, we find that reasoning fine-tuning degrades abstention (by $24%$ on average), even for math and science domains on which reasoning models are explicitly trained. We find that while a carefully crafted system prompt can boost abstention in practice, it does not resolve models' fundamental inability to reason about uncertainty. We release AbstentionBench to foster research into advancing LLM reliability.

Problem

• LLMs struggle with abstaining on unanswerable questions
• No systematic framework evaluates LLM abstention capabilities
• Reasoning fine-tuning worsens abstention despite improving problem-solving

Innovation

• Introduces AbstentionBench for evaluating LLM abstention
• Shows reasoning fine-tuning degrades abstention performance
• Reveals system prompts don't fix uncertainty reasoning