UC Santa Barbara's Group-Evolving Agents hit 71% on SWE-bench by sharing experiences and modifying their own code. A new paradigm for agentic AI.
Advertisement