AI RESEARCH
Auto-Discovery-Bench: Diagnosing Structured State Tracking in Oracle-Guided Discovery
arXiv CS.AI
•
ArXi:2502.15224v2 Announce Type: replace-cross Interactive discovery requires agents to maintain and update structured beliefs over many rounds of feedback. Before evaluating agents in noisy, open-ended scientific environments, it is useful to isolate this prerequisite capability under controlled conditions. We