mlsanityData to 14 Oct 2026

Can a Model Catch Its Own Hallucinations for Free?: Label-Free Doubt Signals Hold Their Own Against a Labelled Dataset for Abstention

arXiv 2608.26121 ↗First posted 19 Jun 2026cs.CL cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Large language models state false facts as fluently as true ones, yet a model often "knows" internally when it is on shaky ground: the probability it assigns to its own answer tends to dip on the facts it gets wrong. The usual way to act on this, teaching a model to abstain rather than guess, requires a labelled dataset of right and wrong answers. We ask whether the model's own confidence, which is free and needs no labels, can do that job instead. We fine-tune each model (with LoRA) to answer when its frozen confidence is high and to say "I'm not sure" when it is low, using the signal alone and no correctness labels. Across six open-weights models (1B-8B, two families) on short-form factual question answering, with correctness adjudicated by an independent judge model, this label-free recipe holds its own against label-supervised abstention-tuning: at matched coverage we find no statistically detectable difference between the two. A control that drills hard examples instead of abstaining does not help, indicating the gain comes from calibration, not rote memorization. The signal's one blind spot is confidently wrong facts, which it cannot flag. A model's own doubt is thus a near-free substitute for a labelled dataset when teaching it when to abstain. Code and artifacts are available on request.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 23 of its 25 references that are in the map. First posted on arXiv 19 Jun 2026.

Builds on

23 of 25 references.

References · 23

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  3. 2025Qwen3 Technical Reportcited by 8,820
  4. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,955
  5. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,998
  6. 2022When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memoriescited by 1,128
  7. 2023R-Tuning: Instructing Large Language Models to Say `I Don't Know'cited by 111
  8. 2024Learning to Route LLMs with Confidence Tokenscited by 67
  9. 2024Factual Confidence of LLMs: on Reliability and Robustness of Current Estimatorscited by 61
  10. 2024Finetuning Language Models to Emit Linguistic Expressions of Uncertaintycited by 26
  11. 2025Robust Hallucination Detection in LLMs via Adaptive Token Selectioncited by 19
  12. 2025Mitigating LLM Hallucination via Behaviorally Calibrated Reinforcement Learningcited by 18
11 more
  1. 2025MetaFaith: Faithful Natural Language Uncertainty Expression in LLMscited by 16
  2. 2024Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Modelscited by 14
  3. 2025Uncertainty Distillation: Teaching Language Models to Express Semantic Confidencecited by 13
  4. 2025Gatekeeper: Improving Model Cascades Through Confidence Tuningcited by 10
  5. 2026Know When You're Wrong: Aligning Confidence with Correctness for LLM Error Detectioncited by 8
  6. 2026Why Fine-Tuning Encourages Hallucinations and How to Fix Itcited by 7
  7. 2024Enhancing Trust in Large Language Models via Uncertainty-Calibrated Fine-Tuningcited by 5
  8. 2026Geometry-Calibrated Conformal Abstention for Language Modelscited by 5
  9. 2026How LLMs Detect and Correct Their Own Errors: The Role of Internal Confidence Signalscited by 3
  10. 2025Detecting AI Hallucinations in Finance: An Information-Theoretic Method Cuts Hallucination Rate by 92%cited by 2
  11. 2026KARL: Mitigating Hallucinations in LLMs via Knowledge-Boundary-Aware Reinforcement Learningcited by 2

Built on by

Not cited yet.