Beyond time for Apple to go in new direction with AI leadership: Big Technology's Alex Kantrowitz

Key Concepts

• Apple Intelligence: Apple's promised AI capabilities, aiming to leverage user data for personalized assistance while prioritizing privacy.
• Large Language Models (LLMs): Advanced AI models that power chatbots and other generative AI applications.
• Siri: Apple's virtual assistant, which has faced criticism for its performance and delayed updates.
• Gemini: Google's LLM, integrated with Apple's AI efforts.
• ChatGPT: OpenAI's LLM, considered a benchmark in the field.
• Privacy-Aware AI: AI systems designed to protect user data and maintain confidentiality.
• Commoditization of AI: The idea that AI technologies will become increasingly similar and interchangeable, reducing differentiation.
• In-house Development vs. Integration: The strategic decision between building AI models internally or integrating third-party models.
• Vulnerability: A weakness or potential risk in a company's strategy or operations.

Apple's AI Strategy and Challenges

The discussion centers on Apple's approach to Artificial Intelligence (AI), particularly in light of perceived shortcomings in its current offerings, such as Siri. The premise is that Apple has not "gotten AI right" in some respects.

1. What Success Would Have Looked Like for Apple:

• Delivering on "Apple Intelligence" Promise: The core expectation was to fulfill the promise of "Apple Intelligence" made about 18 months prior. This involves making sense of the vast amount of data users generate on their iPhones in a privacy-aware manner, a capability that many other companies lack.
• Basic Functionality: Even basic functionalities, like Siri accurately cross-referencing email to confirm flight status (e.g., "When's my flight coming in?" or "Is my flight delayed?"), have not been consistently delivered. This failure to execute on fundamental tasks is identified as a primary reason for Apple's current predicament.
• No "Arms Race" for Core LLMs: Apple is not aiming to compete in the development of core Large Language Models (LLMs) themselves, suggesting a different strategic focus.

2. Recent Changes and Their Direction:

• Perceived Capitulation: The recent moves are interpreted as a "capitulation," indicating a significant scaling down of ambitions.
• New Head of AI: The appointment of A.M. Subramanian as the new head of AI is highlighted. Subramanian has a substantial background, having spent 16 years at Google, including eight as a research scientist.
• Focus on Integration: Subramanian's role is expected to be managing the integration of Siri with Google's Gemini. This suggests a reliance on external models rather than in-house development for the most critical AI components.
• Avoiding Mishaps: The emphasis is on ensuring a smooth integration with Gemini for the upcoming Siri release, as another "mishap" is something Apple "can't afford."
• Strategic Reliance on Google: Apple is not planning to develop the most important AI models in-house. Instead, they are prioritizing seamless integration with Google's models.

3. Long-Term Viability and Perspectives:

• The "Commoditization" Argument: One perspective suggests that AI technology will become commoditized. As LLMs like ChatGPT and Gemini become increasingly similar, Apple might benefit from "sitting it out" and leveraging its existing device ecosystem.
• Device Ecosystem Advantage: Apple's strength lies in its vast user base and the devices through which people interact with technology. If AI becomes a ubiquitous layer, Apple could potentially "take its toll" on that activity by being the primary interface.
• Loss of Control: A significant counter-argument is that by outsourcing model development, Apple gives up a substantial amount of control. If AI proves to be a "sea change" technology, relying on others' models could put Apple in a reactive, "operating from behind" position.
• Vulnerability: This reliance on external models is identified as a "vulnerability," though its severity will be determined by time. The initial strategy was to build in-house, but a change in direction has occurred.

4. Comparison of LLMs:

• Google's Gemini vs. OpenAI's ChatGPT:
  • Gemini: Described as a "good product" but has "faults." It is perceived as being "very safety-oriented" and "thinks a lot."
  • ChatGPT: Considered a "better product," developed by OpenAI. It is noted for its ability to "do a little bit more" and "go out on a limb," suggesting greater flexibility and perhaps less constraint compared to Gemini.
• Google's Credit: There's a suggestion that Google might be receiving "a little too much credit" for being in the lead, implying that ChatGPT, despite being from a startup, has demonstrated superior capabilities.

Conclusion

Apple's current AI strategy appears to be a pragmatic shift towards integrating existing, advanced LLMs, particularly Google's Gemini, rather than developing its own core models. This approach aims to leverage Apple's strong user base and privacy-focused ecosystem while mitigating the risks and costs associated with cutting-edge AI development. However, this reliance on third-party models introduces a significant long-term vulnerability, as it cedes control over a potentially transformative technology. The success of this strategy hinges on the effective integration of these external models and the continued relevance of Apple's platform in an increasingly AI-driven landscape. The comparison between Gemini and ChatGPT suggests that while Google's offering is competent, OpenAI's ChatGPT may currently hold an edge in terms of innovation and flexibility.