Reinforcement Learning
Training by trial, error and reward rather than by examples. It is how models are tuned to be helpful and to refuse things they should refuse.
Related terms
Training Data
The text, images or code a model learned from. It shapes everything the model knows and every bias it carries, and it is why what your website said two years ago may still be what an assistant repeats.
Function Calling
When a model, instead of answering from memory, asks to run a specific tool — look up a price, check stock — and uses the result. It is what turns a chat assistant into something that can act.
Review Schema
Markup for ratings and reviews, including who wrote them. It feeds the star ratings in search results, and gives an assistant something concrete to cite about your reputation.
Guardrails
The limits placed on what an AI may say or do — refusing certain requests, staying on topic, never inventing a price. Essential on anything customer-facing.
Explainability
How well a system can show why it reached a decision. It matters most where the decision affects somebody — a loan, a diagnosis, a job application — and “the model said so” is not an answer.
Knowledge Graph
A map of things and how they relate — this company, its founder, its address, its services. Search engines and AI systems use one to know that two mentions of a name refer to the same business.