01 / The Model
Fine-tuning
Training an existing model further on your own examples so it follows a style or task more reliably. Useful for consistency, rarely the first thing to try.
How you’ll hear it
“Should we fine-tune it on our reports?”
“Try better instructions and retrieval first. Fine-tuning is the expensive option.”
Avoid
Reaching for fine-tuning to teach the model facts. Retrieval does that better.
The longer version
Fine-tuning adjusts a model’s parameters with extra training on your examples. It is good at making behaviour consistent: format, tone, classification. It is poor at adding knowledge that changes, which is better supplied through retrieval, and it brings real cost in data, time and upkeep.