All terms
Foundations
Foundation Model
A large general-purpose model trained on broad data and adapted to many tasks.
Definition
A foundation model is a large neural network trained on a broad, diverse dataset, usually with self-supervised objectives, so it can be reused across many tasks through prompting or light fine-tuning rather than being built for one. It learns general representations adaptable with little task-specific data. Large language models and large multimodal models are the main examples. The term was coined to highlight both the versatility and the risks of building so much on a shared base.