The authors introduce Multi-Agent Self-Supervision (MASS), an iterative framework for recursive self-improvement.
Multi-agent workflows enable models to recursively improve on open-ended tasks
By alternating workflow optimization and fine-tuning on multi-agent trajectories, a model can self-supervise beyond its individual capabilities.
AI Startup
Hyunin Lee · Jinglue Xu · Jeffrey Seely · Donghyun Lee · Somayeh Sojoudi · Matei Zaharia · +1 more
UC Berkeley · Sakana AI
Research Digest··2 min read
The authors propose Multi-Agent Self-Supervision (MASS), a method where a single model generates, evaluates, and optimizes multi-agent workflows, then fine-tunes on those trajectories.
Why this paper
From Sakana AI and UC Berkeley
In one line
Multi-agent self-supervision enables recursive self-improvement of language models on open-ended tasks, yielding 1.2-1.6x higher performance per token.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (params 27B)
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§