Glossary
Foundation model
A large model pretrained on broad data and adaptable to many downstream tasks through fine-tuning or prompting — the pattern behind modern text, speech, and video models.
Large language models, speech models such as Whisper, and video generators all follow the pattern: one enormous pretraining run, then many cheap adaptations. Data demand therefore concentrates in pretraining and alignment, where volumes are largest. The term comes from a 2021 Stanford report that named the model class.
Why it matters
Foundation-model builders are the principal buyers in large-scale data licensing.