Paper summary
Paper summary: A Survey of Context Engineering for Large Language Models
A survey of over 1,400 papers that defines "context engineering": retrieving, processing and managing everything put into an LLM's context, and how RAG, memory, tools and multi-agent systems build on it.
- Context engineering covers everything that goes into the context: retrieval, generation, processing and management, not just the prompt wording.
- The survey finds an asymmetry: models understand complex contexts far better than they generate equally sophisticated long outputs.