AI Paper Atlas

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

GAUSS: Graph-Assisted Uncertainty Quantification using Structure and Semantics for Long-Form Generation in LLMs

Karthik Somayaji NS, Yuxuan Yin, Peng Li

ICML 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
use GAUSS graph alignment instead of Gen-Binaryfalcon-7b-instruct · BiosSC: -0.1535 -> -0.4118Improved
use GAUSS graph alignment instead of Gen-Binaryllama3-8b-instruct · BiosSC: -0.5986 -> -0.7066Improved
use GAUSS graph alignment instead of Gen-Binaryqwen2-57b-instruct · LongFactSC: -0.3328 -> -0.4226Improved
use GAUSS graph alignment instead of Gen-Binaryfalcon-7b-instruct · WildHalluSC: -0.7628 -> -0.7565No clear effect
use GAUSS graph alignment instead of Gen-Binaryfalcon-7b-instruct · LongFactSC: -0.6443 -> -0.6555No clear effect
use GAUSS graph alignment instead of Centralityqwen2-57b-instruct · WildHalluSC: -0.3728 -> -0.6852Improved
use GAUSS instead of LUQmistral-7b-instruct · BiosSC: 0.0541 -> -0.6643Improved
add structural cost (alpha above 0) to semantic-only alignment (ablation)GAUSS · Bios, LongFact, WildHallu (5 LLMs)SC, PCImproved
add semantic cost (alpha below 1) to structural-only alignment (ablation)GAUSS · Bios, LongFact, WildHallu (5 LLMs)SC, PCImproved

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