The article discusses the development of a large local AI model by PrismML, which has managed to run a Chinese open-source model called Qwen 3.6 with 27 billion parameters on an iPhone 17 Pro. This achievement challenges traditional cloud-based AI models by demonstrating that complex tasks like chat responses, reasoning, and autonomous agents can be handled locally on devices. PrismML’s CEO, Baba Hassibi, predicts that within three years, up to 95% of AI processing could occur locally on smartphones, laptops, and appliances, reducing reliance on cloud infrastructure. The technology used by PrismML involves mathematical tricks beyond standard quantization, allowing the model to be compressed significantly without quality loss. The startup, a spin-off of Caltech, is currently in early discussions with Apple to potentially integrate its technology into future products.
Bias read (Center): The article presents a technological advancement without overt ideological framing. It focuses on technical achievements, industry implications, and potential future developments without taking a clear stance on political or social issues. The tone remains neutral, discussing both the benefits and局限






