AI tools: RAG vs fine-tuning — picking the right adaptation

Retrieval-augmented generation and fine-tuning solve different problems. RAG injects fresh facts at query time. Fine-tuning shapes how the model responds. Picking the wrong tool burns weeks.

What AI does well here

• Use retrieved context when it's well-formatted in the prompt
• Acquire stylistic and structural patterns from fine-tuning data
• Combine both when the use case genuinely needs both

What AI cannot do

• Get fresh facts from fine-tuning alone
• Get consistent voice from RAG alone
• Replace good evals with either technique

Chunks too small lose context; chunks too large dilute embeddings. The right size depends on your documents and queries, not a global default.

What AI does well here

• Embed and retrieve chunks of any size you specify.
• Return the top-k matches for a query.

What AI cannot do

• Pick the right chunk size for your corpus automatically.
• Recover meaning that was split across an unfortunate boundary.

Modern AI tools let you create a personal assistant grounded in your documents without writing any code.

What AI does well here

• Cite the specific file and page when answering from your docs.
• Stay scoped to uploaded sources when asked.
• Surface obscure passages relevant to questions.
• Refresh when you re-upload edited versions.

What AI cannot do

• Update automatically when source files change externally.
• Match a custom-built RAG pipeline on retrieval precision.