AI Paper Atlas

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

Don’t Pass@$k$: A Bayesian Framework for Large Language Model Evaluation

Mohsen Hariri, Amirhossein Samandar, Michael Hinczewski, Vipin Chaudhary

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 12 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 Bayes@N instead of Pass@k, Pass^k, G-Pass@k, mG-Pass@kLLM ranking · biased-coin simulation (11 mimics)Kendall tau vs true rankingImproved
use Bayes@N instead of Pass@k variantsLLM ranking · HMMT'25 (11 LLMs)mean convergence@n trials: 69.5 -> 44.2Improved
use Bayes@N instead of Pass@k variantsLLM ranking · BrUMO'25 (11 LLMs)mean convergence@n trials: 48.5 -> 27.1Improved
use Bayes@N instead of Pass@k variantsLLM ranking · AIME'25 (11 LLMs)convergence within 80 trialsMixed
add non-uniform prior from D=1 to 4 trials of original model (ablation)Bayes@N · synthetic updated-model rankingsKendall tau at small NImproved
use large prior weight D=8 or 16 (ablation)Bayes@N · synthetic updated-model rankingsKendall tau at larger NHurt
add 95% credible interval to ranking (ablation)Bayes@80 · AIME'25, AIME'24, HMMT'25, BrUMO'25rank resolutionMixed
increase number of models ranked to 20 (ablation)Avg@N and Pass@k family · four math datasetsranking convergenceHurt

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