What is Training data?
Training data is all the examples an AI model was trained on: the texts, images or numbers it learned from.
A model is exactly as good as the examples it learned from. If the training data held mistakes or a one-sided picture, the model takes those on. Many models are trained on large parts of the internet. Whatever is common there, the model treats as 'normal'.
It explains bias and blind spots. And it is why you should never casually put sensitive data into a free tool: it can become training data itself.
Put precisely
Training data is not checked for truth. It is mostly chosen for size and availability, so it contains mistakes and one-sided views.
Common misconception
That a model has only seen reliable sources. Large parts come from ordinary internet pages.
Related words
Understand AI, calmly
This word is a start. The inklaretaal Learning curve explains AI step by step in plain language, tuned to your field. The first month is free.