Apple Core AI Framework (developer.apple.com)

Article summary

The article discusses Apple's new Core AI framework, which appears to be a replacement for the previous CoreML API, allowing for the conversion of PyTorch models to run on CPU, GPU, and Apple's Neural Engine. The framework is designed for neural networks and transformers, while CoreML will focus on classic, non-neural machine learning. The new framework requires OS 27+, making CoreML still useful for backwards compatibility. The discussion revolves around the implications of this change and the benefits of on-device AI models.

Main themes

• Core AI framework
• On-device AI models
• Neural networks and transformers
• CoreML and backwards compatibility
• Apple's Neural Engine
• Machine learning optimization

What commenters say

• The new Core AI framework is a replacement for CoreML, with CoreML now focusing on classic, non-neural machine learning.
• On-device AI models are more appealing than cloud-based models due to their potential for better performance and privacy.
• The limits of scaling in AI models have been reached, and smaller models can achieve similar results to larger ones.
• Larger AI models are still more effective, and the benefits of smaller models are not yet clear.
• Apple's move towards on-device AI models will disrupt the current AI industry, which relies heavily on cloud-based models.
• The new framework's requirement of OS 27+ will leave older devices without access to the latest AI features.
• On-device AI models can provide better performance and efficiency, but may also lead to increased battery drain if not implemented carefully.
• The distinction between Core AI, CoreML, and MLX is not clear, and Apple should provide more information on their differences and use cases.