AI Paper Atlas

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

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Uncertainty quantification for LLMs

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 changedWhereMeasuredEffect
replace SCP with CoFact density-ratio reweightingSCP · MedLFQA (Lin shift)Factuality: 0.811 -> 0.895Improved
replace SCP with CoFact density-ratio reweightingSCP · WikiData (Lin shift)Factuality: 0.884 -> 0.896Improved
use CoFact instead of conditional conformal predictionCondCP · MedLFQA (Lin shift)Claims Retained: 0.389 -> 0.715Improved
use CoFact instead of conditional conformal predictionCondCP · WikiData (Lin shift)Factuality: 0.724 -> 0.896Improved
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 -> 0.849Improved
use online CoFact instead of oracle Total baselineTibshirani Total · MedLFQA (Lin shift)Factuality: 0.849 -> 0.895Improved
use online CoFact instead of oracle Total baselineTibshirani Total · WikiData (Lin shift)Claims Retained: 0.768 -> 0.748Hurt
increase ACI step size gamma from 0.001 to 0.01 (ablation)ACI-Simple · MedLFQA (Lin shift)Claims Retained: 0.635 -> 0.131Hurt
use CoFact instead of ACI-Simple gamma 0.005ACI-Simple · MedLFQA (Lin shift)Claims Retained: 0.207 -> 0.715Improved
use CoFact instead of rACI with 5% label noiserACI · MedLFQA (Lin shift)Claims Retained: 0.343 -> 0.715Improved
increase label noise ratio in rACI from 0.05 to 0.20 (ablation)rACI · WikiData (Lin shift)Factuality: 0.895 -> 0.880Hurt
reduce calibration fraction from 0.3 to 0.05 (ablation)CoFact · MedLFQA (Lin shift)Claims Retained: 0.722 -> 0.613Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy