Fine-tuning
Retraining a model a little on your own examples so it picks up a style or skill it didn’t ship with.
Think of it like
Sending a strong generalist off to a short apprenticeship in your specific shop.
Example
Training on 10,000 of your past support replies so the model sounds like your team, not a robot.
How it actually works
Techniques like LoRA make it cheap by nudging a small set of weights instead of all of them. It’s cooling as a first move: most teams now try prompting and RAG first, because a fine-tune is a maintenance burden that goes stale with the base model.
For product teams
Teaching the model your style with examples — powerful but a maintenance cost; try prompting/RAG first.
For engineers
Gradient updates on a base model (often LoRA/PEFT) over labeled examples to shift behavior.
Related
- RAG — Often reached for when RAG would have done the job.
- Prompt engineering — A heavier hammer than good prompt engineering.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome