mlsanityData to 14 Oct 2026

The Confidence Paradox: Can LLM Know When It's Wrong

arXiv 2506.23464 ↗First posted 30 Jun 2025cs.AIcited by 11

Not on the map. Why?

Abstract (arXiv)

Document Visual Question Answering (DocVQA) models often produce overconfident or ethically misaligned responses, especially under uncertainty. Existing models like LayoutLMv3, UDOP, and DONUT focus on accuracy but lack ethical calibration. We propose HonestVQA, a model-agnostic, self-supervised framework that aligns model confidence with correctness using weighted loss and contrastive learning. We introduce two new metrics Honesty Score (H-Score) and Ethical Confidence Index (ECI)-to evaluate ethical alignment. HonestVQA improves accuracy and F1 by up to 4.3% across SpDocVQA, InfographicsVQA, and SROIE datasets, while reducing overconfidence. It also generalizes well across domains, achieving 78.9% accuracy and 76.1% F1-score.

Key facts. Cited by 11 papers in the map (data through 14 Oct 2026). Builds on 15 of its 20 references that are in the map. First posted on arXiv 30 Jun 2025.

Builds on

15 of 20 references.

References · 15

  1. 2021CLIP-Adapter: Better Vision-Language Models with Feature Adapterscited by 1,026
  2. 2019When Does Label Smoothing Help?cited by 988
  3. 2023DocLLM: A layout-aware generative language model for multimodal document understandingcited by 116
  4. 2023On the Calibration of Large Language Models and Alignmentcited by 73
  5. 2023Ethical Reasoning over Moral Alignment: A Case and Framework for In-Context Ethical Policies in LLMscited by 64
  6. 2021Understanding Softmax Confidence and Uncertaintycited by 63
  7. 2023Making Large Language Models Better Reasoners with Alignmentcited by 63
  8. 2024Calibrating Language Models with Adaptive Temperature Scalingcited by 58
  9. 2023Adaptation with Self-Evaluation to Improve Selective Prediction in LLMscited by 50
  10. 2024Enhancing Visual Document Understanding with Contrastive Learning in Large Visual-Language Modelscited by 30
  11. 2024Advancing Multimodal Large Language Models in Chart Question Answering with Visualization-Referenced Instruction Tuningcited by 27
  12. 2024LayoutLLM: Large Language Model Instruction Tuning for Visually Rich Document Understandingcited by 24
3 more
  1. 2020Document Visual Question Answering Challenge 2020cited by 20
  2. 2023Large Language Models for Failure Mode Classification: An Investigationcited by 6
  3. 2025DocVideoQA: Towards Comprehensive Understanding of Document-Centric Videos through Question Answeringcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 4; 2026: 7.

11 papers.

  1. 2025Failure Modes in LLM Systems: A System-Level Taxonomy for Reliable AI Applicationscited by 10
  2. 2025Rewarding the Journey, Not Just the Destination: A Composite Path and Answer Self-Scoring Reward Mechanism for Test-Time Reinforcement Learningcited by 3
  3. 2025FalseCrashReducer: Mitigating False Positive Crashes in OSS-Fuzz-Gen Using Agentic AIcited by 2
  4. 2026Do Large Language Models Know What They Don't Know II? A Fully Behavioral, Non-Cognitive Measure of Epistemic Honestycited by 2
  5. 2025Do LLMs Know They Are Being Tested? Evaluation Awareness and Incentive-Sensitive Failures in GPT-OSS-20Bcited by 1
  6. 2026Generative AI for Managerial Decision-Making under Ambiguity and Sycophancycited by 1
  7. 2026Confidence Calibration in Large Language Modelscited by 1
  8. 2026Are Two LLMs Better Than One? A Student-Teacher Dual-Head LLMs Architecture for Pharmaceutical Content Optimizationcited by 0
  9. 2026LLM-Guided Semantic Bootstrapping for Interpretable Text Classification with Tsetlin Machinescited by 0
  10. 2026When Confidence Fails: Overconfidence in LLMs under Uncertainty and Missing Clinical Informationcited by 0
  11. 2026TRACE: Trustworthy Retrieval-Augmented Conversational Enginecited by 0