RealFeedApp beta
- MASS improves score per output token on research benchmarks
The MASS system optimizes multi-agent workflows and distills collective intelligence to serve as a self-supervision signal, improving scores per output token on research benchmarks.
- Field
- training language models
- What they did
- The MASS system optimizes workflows of a group of artificial agents, distills collective intelligence from them, and uses it as a self-supervision signal, which improves the model's score per output token on research benchmarks.
- Why it matters
- This could help create an AI capable of continual self-learning in a setting where the AI being optimized also acts as the evaluator, potentially advancing beyond the knowledge accumulated by humanity.
#recursive self-improvement#multi-agent systems#collective intelligence#mass#ai safety
