04:00
2026-10-15
arxiv.org
artificial-intelligence
ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence
A new arXiv paper (2610.14381v1) introduces ACED-BENCH, a neuro-symbolic benchmark that tests whether LLM agents can actively investigate a system rather than merely answer prompts. The benchmark's ma…