Language Model Teams as Distrbuted Systems

This paper proposes using distributed systems as a principled framework for creating and evaluating LLM teams, rather than relying on trial-and-error. The authors find that fundamental advantages and challenges from distributed computing also arise in LLM teams, offering practical insights from the cross-pollination of these two fields.