Large Scale AI on Apple Silicon - Algorithmic Improvements and the EXO Framework

Explore how to run large-scale AI workloads on Apple Silicon hardware through this 21-minute conference talk from the AI Engineer World's Fair. Learn about the concept of the "hardware lottery" - how research ideas succeed based on their compatibility with current hardware rather than universal superiority - and discover recent algorithmic improvements specifically designed for Apple's silicon architecture. Understand how the EXO Framework enables inference, fine-tuning, and training of large machine learning models on Apple Silicon, scaling from individual MacBooks to clusters of M3 Ultra Mac Studios. Gain insights into distributed systems engineering approaches that challenge the traditional view of hardware as a fixed constraint, and see practical demonstrations of running large ML models efficiently on Apple's ecosystem. The presentation covers the intersection of hardware optimization and machine learning algorithms, showing how breakthrough performance can emerge from better hardware-software alignment rather than purely algorithmic advances.