Training data
학습 데이터
The text, images, sound and other material shown to an AI when it is taught. What it learned from decides how good it is.
Overview
An AI model looks at training data and picks up the patterns and rules in it. Chat AI sees huge amounts of books, web pages and code, and image AI sees huge amounts of images paired with their descriptions. A collected and organized set of such material is called a dataset.
Quality matters as much as quantity. If the data leans one way, the AI's answers lean that way too. Whether other people's writing and pictures can be used without permission, and whether personal information got mixed in, are still disputed. Synthetic data, which is made by AI to fill gaps in the material, is also used.
An everyday example
It is like how what a student knows and does not know depends on which textbooks and workbooks they studied from. If you learn from a wrong textbook, you learn wrong.
Why it matters
Copyright disputes between AI companies and news publishers and authors, and news of data licensing deals, are all stories about training data. Why an AI does not know recent events also depends on the date up to which its training data runs.
Also calleddataset