AI Paper Atlas

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

Efficient Inference for Noisy LLM-as-a-Judge Evaluation

Yiqun T. Chen, Sizhu Lu, Sijia Li, Moran Guo, Shengyi Li

ICML 2026 · Large language models › Language, safety and interpretability · approach: Uncertainty quantification for LLMs · 0 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
use EIF/PPI++ instead of standard PPIPPI · Binary simulation (q0=q1)CI width: 35--55% narrowerImproved
use PPI/EIF instead of Rogan-GladenRogan-Gladen · Binary simulation (q0=q1)CI width: 3--15x narrowerImproved
add debiasing (RG, PPI, PPI++, EIF) to naive judge meannaive LLM-judge mean · Binary simulationcoverage of 90% CIImproved
use Rogan-Gladen with small calibration set and weak judgeRogan-Gladen · Binary simulation (small calibration, low judge quality)coverage of 90% CIHurt
use flexible model (GAM/spline/per-class) for E[Y|Yhat] instead of linearEIF-linear / PPI++ · Discrete-prediction simulation (mu3 varied)CI widthImproved
use EIF-linear instead of PPI identity mapPPI · Discrete-prediction simulation (mu3 varied)CI widthImproved
use EIF/PPI++/MLE instead of Rogan-GladenRogan-Gladen · Chatbot Arena win rate, 1000 splitsCI width: 0.76--0.96 -> 0.27--0.30Improved
use PPI++/EIF instead of PPIPPI · Chatbot Arena win rate, 1000 splitsCI width: 0.35--0.44 -> 0.27--0.30Improved
replace naive judge mean with debiased estimatorsnaive LLM-judge mean · Chatbot Arena win rate, 1000 splitscoverage of 90% CI: 0% coverage under substantial biasImproved
use labeling budget of only 1% with direct q estimationRogan-Gladen · Binary simulation, 1% labelingRMSE of q1: 0.3 vs 0.1 for q0Hurt

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