Recursive Self-Improvement through Multi-Agent Self-Supervision
2026-10-31
![]()
Recursive Self-Improvement through Multi-Agent Self-Supervision starts where a model's output on open-ended work outruns what anyone can check, leaving the model as the best judge of itself. Its method, MASS, has a single base model propose, run and grade multi-agent workflows, searching over them in an evolutionary loop with structural guardrails, then fine-tunes on its own successful trajectories. Over two cycles on Qwen 3.6-27B it reports 1.2-1.6x better performance per output token on open-ended benchmarks than single-agent approaches, at lower training cost. Sakana AI has a project page, and there is a Hacker News discussion.
Was this useful?