Small frozen LLMs trained over state-filtered symbolic option menus with per-agent LoRA adapters reach competent play in three cooperative spatial games, while behavioral audits show reward and cooperation decouple.
UC Santa Barbara , year=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.MA 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Training Small LLMs as Spatial Multi-Agent Policies
Small frozen LLMs trained over state-filtered symbolic option menus with per-agent LoRA adapters reach competent play in three cooperative spatial games, while behavioral audits show reward and cooperation decouple.