AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Probabilities Are All You Need: A Probability-Only Approach to Uncertainty Estimation in Large Language Models

Manh Nguyen, Sunil Gupta, Hung Le

AAAI 2026 · Large language models › Language, safety and interpretability · approach: Uncertainty quantification for LLMs · 4 citations (OpenAlex)

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 single most-likely generation (NLL) with PRO top-K approximationNLL · TriviaQA, Llama2-13BAUROC: 0.684 -> 0.802Improved
replace NLL with PRO (ablation)NLL · NQAUROC: +0.3% averageImproved
PRO versus semantic density (SD)SD · TriviaQAAUROCNo clear effect
PRO versus semantic density (SD)SD · NQAUROC: +6.5%Improved
use adaptive threshold alpha instead of fixed K=1 (ablation)PRO · TriviaQA, Gemma-2BAUROC: 0.806 -> 0.819Improved
use adaptive threshold alpha instead of best fixed K (ablation)PRO · TriviaQA, Llama2-13BAUROC: 0.809 (K=8) vs 0.802No clear effect
use fixed K=2 instead of K=1 (ablation)NLL · TriviaQA, Gemma-7BAUROC: 0.812 -> 0.834Improved
larger K up to 10 on fixed top-K (ablation)fixed K · TriviaQA, Gemma-2BAUROC: 0.806 -> 0.788 (K=10)Hurt
PRO versus baselines under varying ROUGE-L correctness thresholds (ablation)PRO · SciQ, Gemma-7BAUROC: 0.5: 0.832 vs NLL 0.823Improved
larger model in same familyuncertainty estimators · TriviaQA/SciQ/NQAUROCMixed

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