Research map / Large language models / Language, safety and interpretability
CoFact: Conformal Factuality Guarantees for Language Models under Distribution Shift
Zirui Hu, Zheng Zhang, Yingjie Wang, Leszek Rutkowski, Dacheng Tao
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 |
|---|---|---|---|
| replace SCP with CoFact density-ratio reweighting | SCP · MedLFQA (Lin shift) | Factuality: 0.811 -> 0.895 | Improved |
| replace SCP with CoFact density-ratio reweighting | SCP · WikiData (Lin shift) | Factuality: 0.884 -> 0.896 | Improved |
| use CoFact instead of conditional conformal prediction | CondCP · MedLFQA (Lin shift) | Claims Retained: 0.389 -> 0.715 | Improved |
| use CoFact instead of conditional conformal prediction | CondCP · WikiData (Lin shift) | Factuality: 0.724 -> 0.896 | Improved |
| give the fixed estimator all test samples (Total) instead of the first 500 steps (Partial) (ablation) | Tibshirani et al. oracle · MedLFQA (Lin shift) | Factuality: 0.819 -> 0.849 | Improved |
| use online CoFact instead of oracle Total baseline | Tibshirani Total · MedLFQA (Lin shift) | Factuality: 0.849 -> 0.895 | Improved |
| use online CoFact instead of oracle Total baseline | Tibshirani Total · WikiData (Lin shift) | Claims Retained: 0.768 -> 0.748 | Hurt |
| increase ACI step size gamma from 0.001 to 0.01 (ablation) | ACI-Simple · MedLFQA (Lin shift) | Claims Retained: 0.635 -> 0.131 | Hurt |
| use CoFact instead of ACI-Simple gamma 0.005 | ACI-Simple · MedLFQA (Lin shift) | Claims Retained: 0.207 -> 0.715 | Improved |
| use CoFact instead of rACI with 5% label noise | rACI · MedLFQA (Lin shift) | Claims Retained: 0.343 -> 0.715 | Improved |
| increase label noise ratio in rACI from 0.05 to 0.20 (ablation) | rACI · WikiData (Lin shift) | Factuality: 0.895 -> 0.880 | Hurt |
| reduce calibration fraction from 0.3 to 0.05 (ablation) | CoFact · MedLFQA (Lin shift) | Claims Retained: 0.722 -> 0.613 | Hurt |
Closest papers by meaning
- Language Models with Conformal Factuality Guarantees (ICML 2024 · similarity 0.89)
- Large language model validity via enhanced conformal prediction methods (NeurIPS 2024 · similarity 0.89)
- Conformal Language Modeling via Posterior Sampling (NeurIPS 2026 · similarity 0.88)
- Conformal Linguistic Calibration: Trading-off between Factuality and Specificity (NeurIPS 2025 · similarity 0.88)
- Statistical Hypothesis Testing for Auditing Robustness in Language Models (ICML 2025 · similarity 0.87)
- FactTest: Factuality Testing in Large Language Models with Finite-Sample and Distribution-Free Guarantees (ICML 2025 · similarity 0.87)
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.
