domain adaptation

A general assistant is a capable amateur in every field. Domain adaptation turns it into something closer to a specialist — a model fluent in legal language, clinical notes, or a particular programming stack — by steeping it in a lot of that domain's text and tasks.

In practice this combines techniques: continued pretraining on domain documents to absorb vocabulary and conventions, then fine-tuning on domain-specific instruction pairs to shape behaviour. The payoff is fewer mistakes on jargon, better recall of domain facts, and outputs that match professional conventions. The risk is narrowing: a heavily adapted model may lose some of its broad general ability, so teams keep watching general benchmarks while they specialise.