Apple's AI isn't a letdown. AI is the letdown

Original Article ↗ Hacker News Discussion ↗

AI vs Apple: Where Is the Failure?

  • Some argue Apple is uniquely failing: Siri has been mediocre for years, “Apple Intelligence” was heavily marketed but is largely missing or underwhelming, and this is seen as part of a pattern (Maps, Vision Pro, Mail search).
  • Others say the article is Apple PR spin: reframing “we’re behind” as “AI itself is disappointing.”
  • Several commenters think both can be true: current LLM tech is overhyped and Apple specifically has stumbled.

Reliability, Accuracy, and Proper Use-Cases

  • Many emphasize that LLMs are fundamentally probabilistic: great at fluent language, bad at guarantees. That makes them poor fits for tasks needing strict correctness, consistency, or brand safety.
  • Concerns: hallucinations, inconsistent support answers, unpredictable behavior when touching personal data or doing actions on a user’s behalf.
  • Others push back: they use ChatGPT/Gemini daily for translation, explanation, coding help, SQL, and see huge productivity gains despite occasional errors.
  • Tension over whether “anything less than 100% accurate is useless” is realistic; some note humans are also frequently wrong.

On-Device vs Cloud and Apple’s Technical Constraints

  • Apple’s privacy stance pushes on-device models and its “Private Cloud Compute,” but devices have limited RAM, forcing small models that perform far worse than large cloud models.
  • At Apple’s user scale, fully cloud-based LLM features would require massive infrastructure; some think this is a core blocker.
  • Others argue on-device ML already works well for narrow tasks and that small, task-specific models can be “good enough.”

Siri, OS Integration, and UX Frustrations

  • Numerous anecdotes of Siri failing at simple tasks: music playback, calendar creation from images, home automation, speech-to-text, and context use.
  • Integration attempts often feel worse than standalone chatbots: Gemini in messaging without message access, OS keyboards injecting irrelevant personal context.
  • Several people want Apple to focus on:
    • A genuinely better Siri or even a total Siri replacement.
    • Strong universal/semantic search across emails, files, and system data.

Hype, Naming, and Broader Perspective on AI

  • Critique of the term “AI”: if this were just called “LLMs” or “neural networks,” expectations would be lower and disappointments smaller.
  • Comparisons to Web 1.0: we’re in the “putting brochures on the web” phase for AI—using it to redo old tasks rather than discovering its native strengths.
  • Split views: some see a transformative technology still in its infancy; others see a Wall-Street-fueled fad jammed into products where users neither need nor want it.