Training Data
The collection of examples an AI learns from before you ever use it. For a chat assistant that means a huge amount of text: books, articles, websites, and more. This body of learning material is called training data, and it shapes what the AI knows, how it writes, and where its blind spots lie.
Before an AI can help you, it has to learn, and it learns from examples. That pile of examples is the training data. For the chat assistants most people use, the training data is an enormous amount of written text gathered from many sources, which the AI studies to pick up patterns of language, facts, and styles of writing.
A useful comparison is a student who has read a vast library. What they can talk about confidently depends on what was on the shelves. If a subject was covered well, they can discuss it fluently. If it barely appeared, or appeared only in outdated form, their knowledge will be thinner and sometimes wrong.
This has real consequences for you. Training data has a cut-off point in time, so an AI may not know about very recent events. It can also carry over mistakes or biases that were present in the original material. And when an AI confidently states something untrue, a hallucination, the roots often trace back to gaps or noise in what it learned from.
Training data is the raw material for the wider machine learning process. Knowing it exists helps you set fair expectations: an AI is knowledgeable, but its knowledge has an edge, and checking important facts yourself is always wise.