Apple gets full Gemini access and uses distillation to build lightweight on-device AI

Editor's take

Apple has secured full access to Google's Gemini models, leveraging distillation techniques to create more efficient, on-device AI capabilities for its devices. This move is significant as it represents a pragmatic approach to AI development, allowing Apple to bypass the costly and time-consuming process of training massive foundation models from scratch. It also highlights a potential shift in how major tech players collaborate in the AI arms race, with Apple seemingly willing to license foundational technology rather than solely rely on proprietary development.

The implications are far-reaching for both consumers and competitors. For users, this could mean more powerful AI features integrated seamlessly into their iPhones and iPads without compromising battery life or requiring constant cloud connectivity. For rivals, it adds another layer of complexity to their own on-device AI strategies and raises questions about the long-term viability of solely in-house LLM development when powerful, distilled models become more accessible.

Future developments to monitor include the performance and feature set of Apple's new on-device AI, particularly how it measures up to existing cloud-based solutions. The extent to which Apple relies on Google's Gemini, and the specific distillation methods employed, will also be crucial. Furthermore, observing whether other major tech firms adopt similar licensing strategies for foundational models, and the resulting impact on the open-source AI community, will be key indicators of the evolving landscape.