Multi-Agent Self-Supervision

Latest news

1. 12 Oct
   Sakana AI and UC Berkeley propose MASS method for recursive self-improvement

   Researchers introduced Multi-Agent Self-Supervision (MASS), where one language model acts as solver, judge and workflow optimizer to improve itself. The paper shows performance gains but warns that shared blind spots can let errors be reinforced across iterations.

   Research · 1 source

   

Articles

No articles about Multi-Agent Self-Supervision yet.