Alignment may be less about “aligning agents” and more about discovering where the agents are. The critical agent may not be the entity that is designed (eg an LLM chat) ,but the hybrid entity resulting from multiple heavily scaffolded LLM loops across multiple systems.
Alignment may be less about “aligning agents” and more about discovering where the agents are. The critical agent may not be the entity that is designed (eg an LLM chat) ,but the hybrid entity resulting from multiple heavily scaffolded LLM loops across multiple systems.