RAG vs Fine-tuning

Short summary

The article clarifies the critical distinction between RAG and fine-tuning: RAG injects knowledge at query time by retrieving relevant document chunks into the prompt, while fine-tuning adjusts model weights to teach consistent behavior and output style. Fine-tuning does not store new facts — it learns token correlations that smear across weights, leading to persistent hallucination on factual queries. The piece includes a comparison table and practical guidance on when to use each approach.

• •RAG changes the prompt with retrieved context; fine-tuning changes the weights for style
• •Fine-tuning cannot reliably teach new facts — it learns correlations, not stored information
• •RAG enables citable sources and easy updates but requires owning a retrieval pipeline

Generated with AI, which can make mistakes.