The Machine Learning Behind Apple Intelligence - Understanding Modern LLM Architecture

Explore a detailed 20-minute technical video breakdown of Apple's Foundation Large Language Models (AFM-server and AFM-on-device) as revealed in their latest technical report. Learn about advanced machine learning concepts including Next Word Prediction with Transformer Decoders, Reinforcement Learning with Human Feedback, Low Rank Adaptation (LoRA), Knowledge Distillation, and Structured Pruning. Dive into technical implementations like Quantization with Palettization and Mirror Descent Policy Optimization with Leave One Out (MDLOO). Through structured chapters covering pretraining, adapters, and RLHF, gain comprehensive insights into the algorithmic advancements powering Apple's language models. Access supplementary materials including write-ups, slides, and notebooks through membership options, while benefiting from relevant research paper references and links to related content about NLP history and transformer architectures.