Schmidhuber Shares Survey on Self-Improving Agents

Jürgen Schmidhuber posted that meta learning and recursive self-improvement are longstanding ideas now revived by foundation models. He linked a new arXiv survey titled Self-Improvements in Modern Agentic Systems. The survey examines how foundation model-based agents adapt through model updates, intrinsic demonstrations, evaluative feedback, extrinsic experience, prompt evolution, memory, tool use, and recursion. Supporting resources include a GitHub list and a survey hub site. The post presents the work as moving from research prototypes toward deployed systems focused on controllable evolution with minimal human oversight.