Records found: 4
'Meta's ARE platform and the Gaia2 benchmark test agents under asynchronous, event-driven conditions to measure proactivity, timing, and collaboration beyond simple search-and-execute.'
'AmbiGraph-Eval introduces a 560-example benchmark that measures LLM performance on ambiguous natural language to Cypher generation. Evaluations of nine models highlight challenges in ambiguity resolution and syntax generation.'
Researchers have developed MMLONGBENCH, a comprehensive benchmark to evaluate long-context vision-language models on diverse tasks and input lengths, uncovering key performance insights and challenges.
Salesforce Research introduces UAEval4RAG, a new benchmark framework that evaluates RAG systems' ability to reject unanswerable queries across diverse categories, enhancing the reliability of AI responses.