AI Paper Atlas

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

HUME: Measuring the Human-Model Performance Gap in Text Embedding Tasks

Adnan El Assadi, Isaac Chung, Roman Solomatin, Niklas Muennighoff, Kenneth Enevoldsen

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Zero-shot machine-text detection · 2 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 best embedding model instead of human annotatorsMTEB tasks · MTEB human-eval subset (26 task-language pairs, overall)average score: 77.6 -> 80.1Improved
use best LLM annotator instead of human annotatorsLLM-as-annotator · 19 task-language pairsaverage score: 81.2 -> 76.1Hurt
use best LLM instead of humans on rerankingMistral-Small-24B · RerankingMAP: 88.3 -> 78.0Hurt
use GPT-5 instead of humans on classificationGPT-5 · Classificationaccuracy: 79.1 -> 78.9No clear effect
use best LLM instead of humans on STSMistral-Small-24B · STSSpearman: 76.5 -> 75.0Hurt
replace human with best model on emotion classificationembedding models · EmotionClassificationaccuracy: 45.8 -> 75.4Improved
replace human with best model on ArXiv clusteringembedding models · ArxivClusteringP2PV-measure: 49.2 -> 84.6Improved
replace human with best model on Arabic sentimentembedding models · MultilingualSentiment (ara)accuracy: 95.0 -> 77.5Hurt
replace human with best model on Russian sentimentembedding models · MultilingualSentiment (rus)accuracy: 92.5 -> 81.2Hurt
replace human with best model on Arabic STSembedding models · STS22 (ara)Spearman: 67.5 -> 40.9Hurt
replace human with best model on rerankingembedding models · Reranking (eng)MAP: 87.2 -> 96.4Improved
use best model instead of human on multilingual vs English tasks (win rate vs best models)humans · MTEB human-eval subsetwin rate: 29% -> 0%Mixed

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