Foundation Model

The term "foundation model" was coined by Stanford's Center for Research on Foundation Models (CRFM) in 2021 to describe large AI models trained on vast amounts of general data that serve as a base for building more specialized applications. GPT-4, Claude, Gemini, and Llama are all foundation models.

What makes foundation models transformative is their emergent generalization: a single model trained on broad internet text can be adapted via prompting, fine-tuning, or RLHF to perform tasks it was never explicitly trained for — from customer support to code generation to medical diagnosis assistance.

In enterprise AI strategy, the choice of foundation model is a critical decision. Factors include model capability, context window size, licensing terms, inference cost, and whether the provider allows fine-tuning. The EU AI Act specifically regulates general-purpose AI models, which align closely with the foundation model definition.