Sahil Shrivastava

Semantic Early-Stopping for Iterative LLM Agent Loops

Sahil Shrivastava

Multi-agent large language model (LLM) loops, for example a Writer that drafts and a Critic that revises, are almost always terminated by a fixed iteration cap (max_iterations). This is a syntactic kill-switch: it is blind to whether the answer is still improving, so it over-spen…Read more