What is the best approach to give a model up-to-date facts?

Question

Accepted Answer

RAG — retrieve current documents and add them to the prompt — RAG pulls fresh documents at query time, so updating knowledge means updating the document store — no retraining. Fine-tuned facts freeze at training time and go stale.

Answer

Fine-tuning — bake the facts into the model's weights once

Answer

Distillation — compress the facts into a smaller student model

Answer

Quantization — lower the precision so more facts fit in memory

What is the best approach to give a model up-to-date facts?

Why this is the answer

More Customizing LLMs: Fine-tuning, LoRA & RAG flashcards