Procedural Knowledge at Scale Improves Reasoning

Introduces a Reasoning Memory RAG framework that retrieves procedural knowledge (how-to steps and reasoning patterns) at scale to improve LLM reasoning. Demonstrates that retrieved procedures act as scaffolding for multi-step reasoning tasks.

By Wu, Sachan, Yih, and Chen (Meta FAIR + UCLA). Wen-tau Yih is a senior researcher at Meta FAIR known for foundational work on retrieval-augmented generation.