Foundation Model

Home Glossary Item Foundation Model
« Back to Glossary Index

Foundation Models in machine learning refers to models that are trained on a large volume of data from the internet, and can be further fine-tuned for specific tasks. These models absorb a broad understanding of the world and can be seen as a foundation or base to develop more specialized models. They serve as a generic model foundation that can be utilized for a multitude of applications, significantly reducing data requirements and training efforts for task-specific models.


Foundation models have been leveraged in a variety of areas, including language, vision, and robotics. A well-known example of a foundation model in natural language processing is GPT-3, developed by OpenAI, which has been trained on a diverse range of internet text. Once trained, these models can generate human-like text and are impressively versatile—they can translate languages, write essays, summarize text, and even answer trivia.

Foundation models are not without their challenges and concerns. They can inadvertently learn and reproduce biases present in the data used to train them. Additionally, their decisions are often hard to interpret, raising transparency and accountability issues. Also, the quality and accessibility of the data needed to train such models can be a significant barrier.

« Back to Glossary Index