Sakana AI proposes MASS: multi-agent self-supervision for recursive self-improvement
paulabartabajo_ · x · 2026-10-14
Sakana AI and UC Berkeley published "MASS: Recursive Self-Improvement through Multi-Agent Self-Supervision," tackling whether frontier models can push past the frontier of accumulated human knowledge.
- The problem: on non-verifiable tasks, even human experts struggle to evaluate AI-generated research proposals, making the model itself the best available optimizer and evaluator.
- The method: MASS repeatedly optimizes multi-agent workflows, then distills the collective intelligence of multi-agent systems into a self-supervision signal for recursive self-improvement (RSI), improving score per output token over the base model on research benchmarks.
- Paper and code are public; authors include UC Berkeley's Hyunin Lee, Somayeh Sojoudi and Matei Zaharia plus Sakana AI's Yujin Tang, among others.
The reposter calls conquering the space of non-verifiable tasks the next big milestone for AI.
More from Research
- Submarine cyber ops study: operator cognitive capacity swings with the tactical environment — J_MaestreVidal · 2026-10-14
- Recova Trains Robots to Self-Recover in Simulation, Cutting Human Intervention From 87.5% to 0% — Lianhuiq · 2026-10-14
- NVIDIA's AlphaNIST Learns Robotic Assembly From Scratch With Only Binary Rewards, Hitting 98.2% Across 20 Tasks — abhishekunique7 · 2026-10-14
- Encrypted Chain-of-Thought Can Be Decoded: Reasoning Theft Attack on Top LLM APIs — matthew_d_green · 2026-10-14
- Most people carry their mother's cells in their brain, even at 80 — AnnaCiaunica · 2026-10-14
- Looping Layers Boosts Small Decision Models: 1.4B Nears 3x Larger Model — rohanpaul_ai · 2026-10-14