Apple Tests Ask Siri Feature

🔑 Enhanced Key Takeaways

• •The 'Ask Siri' feature is reportedly leveraging Apple's proprietary 'Ajax' large language model, marking a shift from the previous template-based intent recognition system.
• •Internal testing suggests the standalone app will prioritize on-device processing for sensitive queries, utilizing the Neural Engine in the latest A-series and M-series chips to maintain privacy standards.
• •This initiative is part of a broader organizational restructuring within Apple's AI/ML division, aimed at accelerating the integration of generative AI across the entire iOS and macOS ecosystem.

🛠️ Technical Deep Dive

• •Architecture: Transitioning from a distributed intent-classification system to a centralized Transformer-based architecture.
• •On-Device Inference: Utilization of 4-bit quantization techniques to run LLM parameters locally on Apple Silicon.
• •Context Window: Implementation of a persistent, encrypted local vector database to maintain conversational state across sessions.
• •Privacy: Integration of Private Cloud Compute (PCC) for queries exceeding local hardware capacity, ensuring data is not stored or accessible by Apple.

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline