Apple Considers Major Reversal to Improve Siri

Key Concepts:

• In-house model building vs. third-party LLMs
• Apple Foundation Models (internal LLM)
• Generative AI and Siri's performance
• On-device AI vs. cloud-based AI (Private Cloud Compute)
• Anthropic and OpenAI as potential partners
• Custom silicon and Apple's cloud infrastructure

1. Apple's Internal Evaluation of LLMs

• Apple is internally evaluating whether to move away from its own "Foundation Models" (internal large language model system).
• The goal is to improve Siri's performance, which has been lacking despite Apple's investment in generative AI.
• The evaluation is driven by the fact that Siri hasn't improved significantly even after a year of announcing AI features.

2. Comparison to Other Tech Companies

• The situation is compared to Amazon, which uses Claude (Anthropic) for the new Alexa, and Samsung, which uses Google Gemini to power its AI system.
• This highlights a trend of tech companies leveraging external LLMs for specific AI functionalities.

3. Market Reaction

• Apple's stock price increased (1.7% today, 2% yesterday), suggesting the market is responding positively to the news of this potential shift.

4. Apple's Request to Anthropic and OpenAI

• Apple has requested Anthropic and OpenAI to develop versions of their models that can run on Apple's cloud infrastructure, which relies on custom silicon.

5. On-Device AI vs. Cloud-Based AI

• On-device AI: LLMs run directly on the iPhone, iPad, or Mac for lower-tier tasks like summarizing emails.
• Cloud-based AI (Private Cloud Compute): Used for more demanding tasks like summarizing long articles or writing blocks of text.
• Apple is considering using Anthropic or OpenAI models in the cloud due to the higher processing power requirements.

6. Privacy Considerations

• Using cloud-based models from third parties is seen as more private.
• It keeps the third-party models sectioned off from user data, rather than having them deeply integrated into the iPhone infrastructure.

7. Evaluation Process and Concurrent Development

• Anthropic and OpenAI are currently training their models for Apple's evaluation.
• Apple is concurrently developing a version of Siri for next spring that uses Apple Foundation Models.
• This means there are competing projects: one using in-house models and the other exploring external LLMs.

8. Conclusion

• The fact that Apple is heavily considering and evaluating external LLMs indicates a significant shift in their AI strategy.
• The final decision remains to be seen, but the evaluation process itself is telling.