What is fine-tuning?

In practice

Fine-tuning needs a few hundred to a few thousand clean, consistent examples, and the work is mostly in producing those, not in the training. It also fixes you to a model version: when the base model is superseded, the tuning has to be redone. Treat it as a commitment with an ongoing cost, not a one-off improvement.

Where it fits in a build

It earns its place when you need a very specific output structure every time, when a smaller cheap model needs to match a larger one on one narrow task, or when the tone must match a house style that prompting cannot pin down.

Common mistakes

• Fine-tuning to add knowledge, which is what retrieval is for.
• Training on inconsistent examples, which teaches the model to be inconsistent.
• No held-back test set, so nobody can prove the tuned model is better than the prompt it replaced.

Related terms

• Prompt engineering
• Retrieval-augmented generation (RAG)
• Large language model (LLM)

Working on something that involves this?

ASTACKRA designs and builds AI systems, automation and custom software for businesses that need technology shaped around their own workflow. If this term turned up in a proposal and you want a straight answer about whether it applies to your situation, ask us — no obligation, and we will tell you if the answer is no.