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

arXiv: https://arxiv.org/abs/2608.26121 · HTML: https://mlsanity.com/paper/2608.26121/

## 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 (most cited first, 23 of 23)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension](https://mlsanity.com/paper/1705.03551/) (2017)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memories](https://mlsanity.com/paper/2212.10511/) (2022)
- [R-Tuning: Instructing Large Language Models to Say `I Don't Know'](https://mlsanity.com/paper/2311.09677/) (2023)
- [Learning to Route LLMs with Confidence Tokens](https://mlsanity.com/paper/2410.13284/) (2024)
- [Factual Confidence of LLMs: on Reliability and Robustness of Current Estimators](https://mlsanity.com/paper/2406.13415/) (2024)
- [Finetuning Language Models to Emit Linguistic Expressions of Uncertainty](https://mlsanity.com/paper/2409.12180/) (2024)
- [Robust Hallucination Detection in LLMs via Adaptive Token Selection](https://mlsanity.com/paper/2504.07863/) (2025)
- [Mitigating LLM Hallucination via Behaviorally Calibrated Reinforcement Learning](https://mlsanity.com/paper/2512.19920/) (2025)
- [MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs](https://mlsanity.com/paper/2505.24858/) (2025)
- [Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models](https://mlsanity.com/paper/2412.06748/) (2024)
- [Uncertainty Distillation: Teaching Language Models to Express Semantic Confidence](https://mlsanity.com/paper/2503.14749/) (2025)
- [Gatekeeper: Improving Model Cascades Through Confidence Tuning](https://mlsanity.com/paper/2502.19335/) (2025)
- [Know When You're Wrong: Aligning Confidence with Correctness for LLM Error Detection](https://mlsanity.com/paper/2603.06604/) (2026)
- [Why Fine-Tuning Encourages Hallucinations and How to Fix It](https://mlsanity.com/paper/2604.15574/) (2026)
- [Enhancing Trust in Large Language Models via Uncertainty-Calibrated Fine-Tuning](https://mlsanity.com/paper/2412.02904/) (2024)
- [Geometry-Calibrated Conformal Abstention for Language Models](https://mlsanity.com/paper/2604.27914/) (2026)
- [How LLMs Detect and Correct Their Own Errors: The Role of Internal Confidence Signals](https://mlsanity.com/paper/2604.22271/) (2026)
- [Detecting AI Hallucinations in Finance: An Information-Theoretic Method Cuts Hallucination Rate by 92%](https://mlsanity.com/paper/2512.03107/) (2025)
- [KARL: Mitigating Hallucinations in LLMs via Knowledge-Boundary-Aware Reinforcement Learning](https://mlsanity.com/paper/2604.22779/) (2026)

## 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.