Apple Finally Admits It Needs Google to Make Siri Less Stupid
Apple has unveiled its Apple Foundation Model update, effectively turning Siri into a functional assistant by offloading the heavy lifting to Google. It’s the tech equivalent of a genius finally asking for the study guide.
The new Apple Foundation Model (AFM) architecture introduces a two-tier system: local processing for basic tasks and cloud-based heavy lifting. Siri now maintains context across multiple requests, generates images on command, and performs real-time web searches, effectively catching up to features that have existed elsewhere for ages.
For local operations, Apple introduced AFM 3 Core, a 3-billion parameter model optimized for Apple Silicon. For more complex needs, the AFM 3 Core Advanced uses an Instruction-Following Pruning (IFP) technique that stores weights in NAND flash memory, activating only necessary 'experts' to keep the phone from melting. When the device hits a wall, it offloads complex logic to Google servers via Private Cloud Compute (PCC), which promises privacy by not storing user data.
The suite also includes AFM 3 Cloud for multimodal tasks and AFM 3 Cloud Pro, which leverages Nvidia GPUs in Google Cloud for high-level reasoning. Voice synthesis and dictation have seen significant quality bumps, with Apple using Quantization Aware Training to keep these models lean enough to run without requiring a cooling tower strapped to your pocket.
It is truly heartwarming to see a trillion-dollar company finally discover the basic utility of a chatbot that actually remembers what you said thirty seconds ago. Apple has masterfully marketed 'arriving at the party two years late' as a revolutionary breakthrough in intelligence, proving once again that if you polish a feature long enough, people will eventually call it premium innovation.
Source: Apple Machine Learning Research
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.