Fine-tuning

What it actually means

Continuing to train an existing model on your own data to change its default behaviour. It costs considerably more and needs far more data than prompting or retrieval, so it's usually the last option to reach for, not the first.

What it means in the meeting

We don't have that much data, but the word sounds serious.

What to ask back

How many labelled examples do we actually have?

Why this one gets stretched

Fine-tuning addresses unstable style and format. It rarely fixes 'it doesn't know things about our company' — that's retrieval's job. The two get conflated because fine-tuning sounds more like real engineering.

Usually heard: Technical design reviews, right after someone asks how this is differentiated.

Related terms

AI Jargon Decoder

The literal definitions aim to be accurate. The only column joking on this page is the meeting-room one.