According to reports, Apple is in talks with the Silicon Valley startup PrismML to develop a revolutionary AI model compression technology. The technology aims to compress large AI models so they can run natively on iPhones. If this technology proves successful, it could enhance Apple's advantage in privacy protection and provide strong support for the currently slow progress of Siri upgrades.

The CEO of PrismML revealed that Apple has begun evaluating the feasibility of its technology. Although the discussions are still in the early stages, the progress looks quite promising. Notably, this news came just one day after Apple released the iOS 27 beta version, which marks the first public test of the significantly revamped Siri, aiming to compete with AI assistants like OpenAI and Anthropic.

By keeping more AI computing tasks on the device, Apple can reduce response latency, lower cloud computing costs, and allow some features to be used without an internet connection. This aligns closely with Apple's long-standing emphasis on privacy protection. Pathak, an analyst at Counterpoint, pointed out that the combination of cloud and device-side AI will provide users with a more comprehensive, efficient, and privacy-focused AI experience: complex tasks are handled in the cloud, while sensitive information is processed on the device.

PrismML was initially incubated at Caltech and received support from Khosla Ventures. The company has already publicly released a compressed version of Alibaba's open-source model Qwen, reducing the original model size from about 54GB to less than 4GB, allowing the model's 27 billion parameters to run on iPhone 15 and newer models. In the future, PrismML plans to handle other large open-source models to further advance this technology.

On the technical level, PrismML's compression solution can reduce a cloud model that originally required eight GPUs to just one, and allows models that rely on data centers to migrate to smartphones and laptops. Although this method effectively reduces the memory and computing power needed for a single AI task, it does not mean that the overall chip demand will decrease.

Key Points:

🔍 Apple is in talks with PrismML to compress large AI models so they can run on iPhones.  

📱 This technology has the potential to enhance privacy protection and support the upgrade of Siri.  

💡 PrismML has successfully compressed a large model to less than 4GB, compatible with iPhone 15 and newer models.