Retrieval-augmented generation

Retrieval-augmented generation is a technique where a language model is given relevant documents to read before it answers, instead of answering from memory.

It is the difference between asking a model what it recalls about a topic and asking it to answer from a specific file in front of it.

For legal work the retrieval half is what matters. An answer is only as good as the documents that were fetched to support it, and an answer that cannot name those documents cannot be checked.

Related: Hallucination, Claim

Legalnaut builds this out of your own case file. See the plans.