SESKStuff Everyone Should Know
SESK: Stuff Everyone Should Know · sesk.info/ai-rag-parametric-plus-retrieval

What two kinds of memory does retrieval-augmented generation (RAG) combine?

The model's own weights plus a searchable document index.

Start playing free

  1. RAM and hard drive storage
  2. Weights plus retrieved documents ✔
  3. Short-term and long-term chat history

Why: Knowledge stored in the weights is parametric memory. RAG adds non-parametric memory: a retriever fetches relevant passages, and the generator writes its answer using them. The original paper used a vector index of Wikipedia.

Huh, didn't know that: RAG makes it easier to update knowledge and show sources: swap the documents instead of retraining the model.

The dad joke
RAG is like an open-book test. The model still has to read the book, though.
Like this card?
Play the game

References

Answer: Weights plus retrieved documents. Knowledge stored in the weights is parametric memory. RAG adds non-parametric memory: a retriever fetches relevant passages, and the generator writes its answer using them. The original paper used a vector index of Wikipedia.

  1. Lewis et al., Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (arXiv 2005.11401) arxiv.org/abs/2005.11401
    models which combine pre-trained parametric and non-parametric memory for language generation
    Checked 2026-10-10.

Think this answer is wrong? Tap "Challenge this answer" on the card in the game and tell us why.

Test yourself in the game

Artificial IntelligenceAt its core, what is a large language model trained to do?Artificial IntelligenceWhat does an LLM's context window measure?Artificial IntelligenceWhere does an indirect prompt injection hide?

That's stuff everyone should know.