AI Paper Atlas

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

ABCD: All Biases Come Disguised

Mateusz Nowak, Xavier Cadet, Peter Chin

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
replace letter labels and letter prediction (S&L) with dash labels and matched full-text answers (M&D)13 LLMs · five MCQ benchmarksaccuracy variance across permutations: variance ratio geometric mean 0.33Improved
use M&D instead of S&L13 LLMs · NonsenseQA, 5-shotmedian accuracy: per-model median reductions 12.5% to 52%Improved
use M&D instead of S&L13 LLMs · CSQAvariance: variance ratio 0.11Improved
use M&D instead of S&LLlama-3.1-8b-instruct · CSQAvariance: variance ratio 1.688Hurt
use M&D instead of S&L13 LLMs · ARCvariance: 11 of 13 models lower varianceImproved
use M&D instead of S&L13 LLMs · GPQAvariance: 9 of 13 models lower varianceImproved
use M&D instead of S&L13 LLMs · MMLU-Provariance: 7 of 13 cases lower variance; ratio 0.99No clear effect
use M&D instead of S&L13 LLMs · INCLUDE (es, fr, it, de)variance: variance ratio 0.23Improved
use M&D instead of S&L13 LLMs · SCORE robustness on MMLU-Pro and ARCSCOREHurt
use M&D instead of S&L when few-shot distribution is unchanged13 LLMs · GPQASCORE: 10 of 13 models improvedImproved
vary sentence similarity model and similarity function (ablation)M&D matching · benchmarksaccuracyNo clear effect
use homogeneous option labels instead of heterogeneous labels (ablation)M&D · benchmarksevaluation stabilityImproved

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