A foundation model is a large AI model trained on a broad mix of data so it can be adapted to many different tasks. Developers then build products on top of it, such as chat assistants, writing helpers and search tools. Think of it like a basic dough recipe: one batch can become bread, pizza or rolls, depending on what the baker does next. Large language models are the best-known kind, and some foundation models also handle images or audio (see multimodal model).
Why it matters
Most AI tools a small organization uses are built on a foundation model from one of a few developers. The chatbot in your donor database and the assistant in your email may share the same base, along with its strengths, gaps and weaknesses. A problem in one foundation model can show up in many products at once. Before you adopt a tool, ask the vendor which foundation model it uses and whether they have adjusted it through fine-tuning. Ask if they can share a model card or similar documentation, and how they will tell you when the underlying model changes.
Published Updated