Apple is in discussions with PrismML, a Silicon Valley startup that has successfully compressed AI models to operate on iPhones, reducing the size of models from approximately 54 GB to under 4 GB. This breakthrough allows all 27 billion parameters of the model to function on devices like the iPhone 15.
PrismML's CEO, Babak Hassibi, indicated that Apple is currently evaluating the technology's speed, energy efficiency, and performance. The discussions are in early stages, and while the outcome is uncertain, progress is being made.
This development is particularly relevant as Apple aims to enhance Siri's competitiveness against AI assistants from OpenAI and Anthropic, while also keeping user data processing on-device to address privacy concerns.
The ability to run more complex AI tasks locally could lead to advancements in features such as computational photography and health monitoring, which rely on sensitive personal data. However, there is a trade-off in performance, as the compressed models may lose some accuracy.
Analysts have noted that while this technology could reduce the need for extensive cloud resources, it does not eliminate the demand for chips entirely; rather, it may shift the demand from data centers to personal devices. The implications for chip demand are significant, as improvements in AI efficiency could alter the landscape of memory chip requirements.
Overall, if successful, this initiative could bolster Apple's AI capabilities while reinforcing its commitment to user privacy and device integration