Foundations
Training data
The text a model learned from — which determines what it knows, and when its knowledge stops.
Models are trained on data up to a cutoff date and know nothing after it. Ask about last week's news and you will get either a refusal or an invention.
Two business implications:
- Currency — anything recent must be supplied at request time via retrieval or a search tool.
- Confidentiality — whether your data is used for training depends on the provider and plan. Business and enterprise tiers generally exclude it; free consumer tiers often do not. This is worth checking explicitly rather than assuming.