Foundation Models
Foundation models are large-scale machine learning models pre-trained on extensive datasets that can be adapted to a wide range of downstream tasks through fine-tuning or prompting. Key characteristics include large parameter counts, general-purpose representational capacity, and support for cross-domain transfer learning. Foundation models encompass language, image, and multimodal types, and serve as core infrastructure for modern AI applications across natural language processing, computer vision, and generative AI.
Core Facts
Timeline (last 90 days)
Apple provides third-party developers with the Foundation Models framework, allowing apps to call on-device language models to complete tasks such as rewriting and error correction, and by default it does not send user data to external servers.
Apple has opened relevant APIs to third-party developers, enabling them to leverage local model capabilities to build features such as context understanding and semantic classification.
All Facts (6)
The term "foundation model" was formally introduced by Stanford University's HAI Institute in 2021.
90%Verified涌现能力（emergent abilities）指模型规模超过某临界点时突然显现训练时未明确优化的新能力，如多步推理、代码生成或少样本学习
75%UnverifiedFoundation Models框架是今年新增的能力，允许应用直接调用Apple的端侧大模型
90%UnverifiedFoundation Models框架将大语言模型推理能力下沉到设备本地执行，而非依赖云端API调用
90%UnverifiedApple provides third-party developers with the Foundation Models framework, allowing apps to call on-device language models to complete tasks such as rewriting and error correction, and by default it does not send user data to external servers.
50%UnverifiedApple has opened relevant APIs to third-party developers, enabling them to leverage local model capabilities to build features such as context understanding and semantic classification.
50%