04:00
2026-10-15
machinebrief.com
large-language-models
PersEvalBench: Can LLM Evaluators Judge Contextually Appropriate Use of Persona Information?
A new benchmark, PersEvalBench, shows that five LLM evaluators struggle to judge whether personalized assistants use user history appropriately, with the strongest evaluator recognizing only 63.9% of …