04:00
2026-08-05
arxiv.org
large-language-models
Every Wrong Answer Counts: Option-Level Psychometrics for LLM Multiple-Choice Benchmarks
Researchers introduced the LLM Nominal Response Model (LLM-NRM), an option-aware psychometric framework that models the full distribution over answer choices to jointly estimate LLM ability and optionβ¦