PARL
Parallel-Agent Reinforcement Learning, Moonshot's training method that teaches orchestrator models to decompose problems and dispatch sub-agents in parallel.
PARL (Parallel-Agent Reinforcement Learning) is Moonshot AI's training methodology for multi-agent systems. Unlike standard agent training that optimizes single-agent task completion, PARL trains an orchestrator model to decompose problems and dispatch sub-agents in parallel while maintaining coherence. The method includes a 'Critical Steps' metric designed to prevent 'serial collapse,' where parallel agents gradually degrade into sequential execution patterns.
Also known as
Parallel-Agent Reinforcement Learning, Parallel Agent RL