Apple reportedly trying to distill Google’s multi-trillion-parameter Gemini AI to run on iPhone

As the integration of generative AI becomes increasingly prevalent in technology, Apple is making significant strides to enhance its Siri virtual assistant. The company has faced delays in rolling out AI improvements since promising an upgraded version of Siri for 2024. However, a strategic collaboration with Google is set to merge Siri with the advanced Gemini AI system later this year, just in time for the upcoming WWDC event. Despite Apple's long-standing commitment to privacy and its preference for local AI processing, recent reports indicate a shift in strategy. According to The Information, the revamped Siri will operate both on-device and in the cloud, which marks a departure from Apple's traditional focus on local processing. This dual approach could raise concerns among Apple enthusiasts who value the company's privacy-centric ethos. With each new chip release, Apple touts its advancements in AI optimization, particularly through its Neural Engine upgrades. However, the reality is that while smartphones are marketed as capable of handling complex AI models, their hardware often falls short. Most mobile GPUs can process more tokens than AI-specific NPUs, yet they still lack the necessary RAM to support vast AI models. The AI models currently running on smartphones are limited in size, typically containing only a few billion parameters. In stark contrast, Google's latest Gemini models boast trillions of parameters. This disparity can lead to on-device AI that operates with reduced precision, making it faster but less accurate in generating responses. Consequently, these smaller models may not deliver the same level of intelligence as their cloud-based counterparts. Google has developed mobile-optimized versions of Gemini, known as Gemini Nano, which are intended for specific tasks like contextual features and audio summarization. However, the requirements of a conversational assistant like Siri demand a different approach, one that is currently reliant on cloud processing. In fact, on Android devices, Google has opted to direct all interactions with Gemini to the cloud, foregoing local processing entirely.