Apple Distills Gemini for Offline Siri

Apple reportedly has the ability to query and edit Google’s Gemini and use Gemini’s outputs and reasoning traces to train much smaller models that run entirely on device for Siri and other features. Those distilled models aim to match Gemini‑level behavior while requiring far less compute and no network connection, though mismatches in Gemini’s tuning (chat/coding) create integration work for Apple. — This matters because it shifts who controls device AI behavior (Apple via distilled models, but relying on Google upstream), with implications for competition, privacy (more offline inference), content control, and supply‑chain concentration in AI.