A foundation model learns general patterns from large and varied data before being adapted or prompted for particular uses. It can support multiple tasks rather than being built for only one narrow prediction.
General-purpose foundation models and specialized systems can progress at the same time. Broad capability is useful, while a focused system may perform better or more efficiently in a specific domain.
ELI5
A foundation model is a broadly trained AI model that can support many different tasks. Developers adapt the same general model with instructions, examples, tools, or additional training instead of building a new model for every feature.
For example, one foundation model might summarize documents, answer questions, and help write code in different applications. Those applications still inherit the model's mistakes and limits, so each use needs its own testing and safeguards.
Why is it called a foundation model?
It provides a reusable base that can support many applications through prompting, tools, fine-tuning or other adaptation.
Is every specialized AI system a foundation model?
No. Some systems are trained or designed for a narrow domain and are not intended as a broad reusable base.
