Google Research Open-Sources RRSI: AI Agents That Improve Their Own Harness Without Overfitting

Google Cloud AI Research and university partners have open-sourced RRSI, a framework enabling LLM agents to self-improve their operational harness without changing model weights, reducing overfitting on benchmarks. RRSI uses regularization techniques to guide iterative enhancements in prompts, tools, and control flow, showing improved performance across multiple benchmarks.