Browse Teach Curate

Multi-Path Reactive-Reflection Agents for LLM Scientific Reasoning (RR-MP)

A training-free framework (RR-MP) that improves LLM agents' scientific reasoning by running multiple parallel reasoning paths, each pairing a reactive agent with a reflection agent that externally stimulates self-correction to prevent Degeneration-of-Thought, then consolidating all paths with a separate summarizer.

Wall-E Text

2501.00430

This paper proposes the Reactive and Reflection agents with Multi-Path Reasoning (RR-MP) framework, a training-free method for improving the scientific-reasoning accuracy of large-language-model agen…