Recursive Self-Improvement and the Future of AI
Researchers claim a working multi-agent system achieving recursive self-improvement across the full AI stack
“You can think of RSI at many different levels of the stack, prompts, harness, weights, and you can go all the way to basically chips. Our goal here is to optimize all the levels of the stack.”
Researchers with backgrounds from MIT Media Lab and DeepMind presented their approach to recursive self-improvement at an a16z event, claiming a multi-agent system that genuinely self-optimizes across the full stack from prompts to chips. They draw a sharp line between true RSI and the many startups that merely use AI to generate artifacts like GPU kernels. The assertion that RSI is not a belief but a demonstrated capability — withheld from release until verified — marks this as a significant claim worth tracking.