NVIDIA's AI Breakthrough: ProRL Agent Decouples LLM Training at Scale

NVIDIA cuts shell command latency from 0.78s to 0.42s with new infrastructure that eliminates training bottlenecks for multi-turn AI agents.