← Concept IndexDEFINITION WHY IT MATTERS COMMONLY CONFUSED WITH
Training data and its limits
Also called: what's in the data, data bias, where knowledge comes from
A model's abilities and blind spots are shaped by the text and images it was trained on — which languages, sources, viewpoints, and time periods were and weren't well represented.
It explains why models are stronger in some languages and topics than others, why they can repeat historical biases, and why they're frozen at a cutoff. The data is the model's whole world; its gaps become the model's gaps.
A neutral, complete view of the world. Training data is a partial, skewed sample, and the model inherits that skew rather than correcting for it.