Research map / Large language models / Language, safety and interpretability
From Sampling to Cognition: Modeling Internal Cognitive Confidence in Language Models for Robust Uncertainty Calibration
Hao Li, Tao He, Jiafeng Liang, Zheng Chu, Ming Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| supervise with CogConf instead of sampling accuracy (ablation) | CogAlign Llama-3-8B · TriviaQA+CommonsenseQA | ECE: 10.2 -> 7.5 | Improved |
| supervise with CogConf instead of sampling accuracy (ablation) | CogAlign Mistral-7B · TriviaQA+CommonsenseQA | AUROC: 70.2 -> 72.7 | Improved |
| supervise with CogConf instead of semantic entropy (ablation) | CogAlign Llama-3-8B · TriviaQA+CommonsenseQA | ECE: 11.5 -> 7.5 | Improved |
| supervise with CogConf instead of semantic entropy (ablation) | CogAlign Mistral-7B · TriviaQA+CommonsenseQA | ECE: 15.4 -> 7.8 | Improved |
| supervise with CogConf instead of sampling accuracy (ablation) | CogAlign Llama-3-8B · SportQA | ECE: 21.3 -> 14.0 | Improved |
| supervise with CogConf instead of semantic entropy (ablation) | CogAlign Mistral-7B · ScienceQA | ECE: 15.6 -> 6.2 | Improved |
| supervise with CogConf instead of sampling accuracy (ablation) | CogAlign · IK vs IDK confidence gap | confidence gap: 1.17 -> 3.70 | Improved |
| supervise with CogConf instead of semantic entropy (ablation) | CogAlign · IK vs IDK confidence gap | confidence gap: 0.72 -> 3.70 | Improved |
| CogAlign instead of RewardingDoubt | calibration RL · Llama-3-8B in-domain | ECE: 9.8 -> 7.5 | Improved |
| CogAlign instead of RewardingDoubt | calibration RL · Mistral-7B in-domain | AUROC: 62.7 -> 72.7 | Improved |
| CogAlign instead of RewardingDoubt | calibration RL · Mistral-7B ARC-Challenge | ECE: 3.2 -> 6.2 | Hurt |
| CogAlign instead of RewardingDoubt | calibration RL · Llama-3-8B SportQA | ECE: 7.5 -> 14.0 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RewardingDoubt — Rewarding Doubt: A Reinforcement Learning Approach to Calibrated Confidence Expression of Large Language Models (ICLR 2026) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs (ICLR 2024 · similarity 0.91)
- Large Language Models Must Be Taught to Know What They Don’t Know (NeurIPS 2024 · similarity 0.90)
- Distinguishing the Knowable from the Unknowable with Language Models (ICML 2024 · similarity 0.89)
- Calibrating Large Language Models with Sample Consistency (AAAI 2025 · similarity 0.89)
- Calibrating Reasoning in Language Models with Internal Consistency (NeurIPS 2024 · similarity 0.89)
- Semantic Density: Uncertainty Quantification for Large Language Models through Confidence Measurement in Semantic Space (NeurIPS 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
