RAG

RAG is a technique that improves AI output accuracy by retrieving relevant documents from a corpus and using them as context for generation. Instead of relying solely on parameters learned during training, the model can reference external sources, reducing hallucination and enabling knowledge that was not in the training data.

The standard RAG pipeline works as follows: a query is embedded into a vector space, similar documents are retrieved based on vector similarity, and the retrieved text is provided to an LLM as context for generating a response. This approach has become widespread for question-answering, search, and knowledge-intensive tasks.

However, RAG has structural limitations for knowledge production. It retrieves text snippets based on similarity, not on argumentative relations. It cannot distinguish support from attack, evidence from opinion, or primary source from restatement. When sources conflict, RAG tends to blend them into smooth summaries rather than surfacing the contention. See RAG Limitations for the full critique.

Coscientist goes beyond RAG by maintaining a Dialectical Graph that stores claims, evidence spans, and typed relations. Retrieval becomes "what bears on this claim?" rather than "what is similar to this query?"