What Are Foundation Models?
Rick Merritt,
NVIDIA Blog,
Jun 22, 2023
As this article reports, "foundation models are AI neural networks trained on massive unlabeled datasets to handle a wide variety of jobs from translating text to analyzing medical images." Because the data is unlabled, foundation models are not steered in any particular direction by developers; they identify patterns in the data wherever they may appear. After training, these models can be applied to more specific and directed tasks, such as answering questions, object recognition, or sentiment analysis. Hence we see in the development of systems like chatGPT a two-part process: first, the development of a foundation model, and then second, the application of that model to a specific task. For more, see this, um, foundational paper from 2021: On the Opportunities and Risks of Foundation Models (214 page PDF).
Today: 6 Total: 88 [Direct link] [Share]
