iPhone 16 Pro Max Neural Engine produces garbage LLM output - 15 Pro runs same code fine
A developer discovered their iPhone 16 Pro Max generates nonsense when running MLX framework LLMs, while an iPhone 15 Pro and MacBook Pro execute identical code perfectly. Tensor outputs show numerical values an order of magnitude wrong, pointing to potential Neural Engine hardware defects in Apple's latest flagship.