mlsanityData to 14 Oct 2026

Can Language Models Learn When to Abstain? Policy Learning and Calibrated Answer Selection

arXiv 2610.13821 ↗First posted 9 Oct 2026cs.CLcited by 0

Not on the map. Why?

Abstract (arXiv)

A useful language model should know when to answer and when to say "I don't know." Can we improve that decision by training the model itself or by training a separate component to check its answers? We study both approaches on Qwen3-14B for factual questions, using a score that rewards correct answers, penalizes mistakes, and allows abstention. Across three training seeds and two held-out confidence thresholds, a learned answer selector improves threshold-averaged utility over the original model's calibrated confidence filter. On TriviaQA, mean coverage increases from 20.2% to 30.9%, while the pooled error rate among scored answers decreases from 23.0% to 14.5% under exact-match grading. The tested reinforcement-learning changes do not establish consistent gains. On NQ-Open, the selector improves threshold-averaged utility over the filter but still performs worse than always abstaining. A partial human-reviewed audit preserves the TriviaQA finding. Our results support learning to select answers in this setting while showing that reliable abstention across datasets remains unresolved.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 22 of its 33 references that are in the map. First posted on arXiv 9 Oct 2026.

Builds on

22 of 33 references.

References · 22

  1. 2025Qwen3 Technical Reportcited by 8,820
  2. 2017On Calibration of Modern Neural Networkscited by 4,960
  3. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  4. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,955
  5. 2022Language Models (Mostly) Know What They Knowcited by 1,856
  6. 2023A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questionscited by 1,601
  7. 2025Understanding R1-Zero-Like Training: A Critical Perspectivecited by 1,345
  8. 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 974
  9. 2017Selective Classification for Deep Neural Networkscited by 852
  10. 2023SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Modelscited by 810
  11. 2022Teaching Models to Express Their Uncertainty in Wordscited by 751
  12. 2019Latent Retrieval for Weakly Supervised Open Domain Question Answeringcited by 695
10 more
  1. 2020Selective Question Answering under Domain Shiftcited by 238
  2. 2025Why Language Models Hallucinatecited by 220
  3. 2025AbstentionBench: Reasoning LLMs Fail on Unanswerable Questionscited by 145
  4. 2023Evaluating Open-Domain Question Answering in the Era of Large Language Modelscited by 111
  5. 2023R-Tuning: Instructing Large Language Models to Say `I Don't Know'cited by 111
  6. 2024SaySelf: Teaching LLMs to Express Confidence with Self-Reflective Rationalescited by 105
  7. 2024Mitigating LLM Hallucinations via Conformal Abstentioncited by 59
  8. 2025TruthRL: Incentivizing Truthful LLMs via Reinforcement Learningcited by 27
  9. 2026I-CALM: Incentivizing Confidence-Aware Abstention for LLM Selective Answeringcited by 5
  10. 2026Rewarding Intellectual Humility Learning When Not To Answer In Large Language Modelscited by 2

Built on by

Not cited yet.