Escher-Loop: Mutual Evolution by Closed-Loop Self-Referential Optimization 事件

Escher-Loop: Mutual Evolution by Closed-Loop Self-Referential Optimization arXiv:2604.23472v2 Announce Type: replace Abstract: While recent autonomous agents demonstrate impressive capabilities, they predominantly rely on manually scripted workflows and handcrafted heuristics, inherently limiting their potential for open-ended improvement. To address this, we propose Escher-Loop, a fully closed-loop framework that operationalizes the mutual evolution of two distinct populations: Task Agents tha