Sakana AI Proposes Self-Improvement Method MASS
Understand the story
According to The Decoder on October 13, 2026, researchers from Sakana AI and UC Berkeley proposed a multi-agent self-supervised method called MASS. The method has a single model, Qwen3.6-27B, take on three roles simultaneously: problem solver, workflow optimizer, and judge, without relying on a stronger teacher model. The report says that with MASS the model's problem-solving and judging capabilities both improved, but this multi-role self-improvement approach also suffers from shared blind spots. The report only discloses the method's introduction and these preliminary results; no further developments have been reported.
AI overview of reporting · Updated 6 days ago
Reporting timeline
Explore different aspects of the story through its reporting.
- The DecoderSakana AI's self-improving model gets better at solving and judging, but can't catch its own blind spots
Researchers at Sakana AI and UC Berkeley propose MASS, a multi-agent self-supervision method that lets a single Qwen3.6-27B model play three roles—solver, workflow optimizer and judge—without needing a stronger teacher model.
Story heat trend
Current heat3·Comparable peak10（10/13 03:00）·Comparable change over 24 hours -49%
The trend compares the same continuously observed sources, which may cover fewer sources than the current heat score. Hover or select the chart to inspect each hour; use the left and right arrow keys to navigate.