Recursive self-improvement through collective intelligence
Researchers at Sakana AI have introduced MASS, a framework for recursive self-improvement where AI models act as their own optimizers and evaluators. This approach aims to overcome the limitations of human supervision in frontier AI development by enabling models to learn and improve autonomously.
Why it matters
As AI capabilities surpass human expertise, developing autonomous self-improvement loops is critical for scaling research and discovery beyond current human-verified boundaries.
To advance beyond humanity's accumulated knowledge, AI itself (the optimizee ) may also be its best available optimizer and evaluator. MASS repeatedly optimizes multi-agent workflows, then distills collective intelligence from multi-agent systems and uses it as a self-supervision signal for RSI. It improves score per output token over the base model on research benchmarks. We hope MASS contributes to AI’s Odyssean moment: continual self-learning in a setting where the optimizee is the evaluator, and the optimizer is the optimizee.
Work done during an internship at Sakana AI.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in