● ResearchNov 2, 2026 · 12:001 source
More Rules, Longer Lists? Benchmarking Self-Evolving Agents on a Hidden-Rule Curriculum
From arxiv.org
AI;DR
CLR-Bench is a Curriculum of Latent Rules benchmark. Each chain adds one hidden rule per level on top of earlier ones. Twenty chains over 10 domains are released for evaluation of six recent self-evolution methods.
Read originalSign in to suggest