Topic · 17 episodes across 9 reviews

Many Models, One System: Collective Dynamics of Multi-Agent LLMs

← all reviews

Two opposite-signed results on what happens when you wire models together — splitting one model across trained agents can double accuracy, but letting models talk freely in a closed loop quietly collapses their ideas.

Covered in these reviews

Related concepts

Episodes and papers for this topic live on the concept pages.

Multi-Agent Systems Emergent Behavior AI Safety Training Methods Agentic Workflows Agentic AI Evaluation & Benchmarks Credit Assignment AI Alignment Agent Memory Reinforcement Learning Ablation Studies Long-Horizon Tasks Agentic Misalignment