A foundation model is a large AI model trained on broad, general data that can be adapted for a wide range of downstream tasks through fine-tuning or prompting.
Foundation models are trained at scale on diverse data — text, images, code, or combinations — and serve as a base that can be specialized for specific applications. Large language models are a prominent type of foundation model.
Because foundation models are trained on broad data, they may reflect a wide range of biases and errors present in that data. Adapting them for specific uses does not automatically remove those issues.