Apple squandered the Holy Grail
Apple’s new “Apple Intelligence” features are widely seen as underwhelming despite strong underlying technology like on‑device models and a tightly designed Private Cloud Compute system for privacy. Commenters argue that many shipped capabilities — from image generation and photo “cleanup” to notification summaries — feel gimmicky, intrusive, or half‑baked, and clash with Apple’s historic “bicycle for the mind” ethos of tools that enhance user agency. Some remain optimistic that Apple will iterate as it did with Maps, while others see the rushed AI push as a strategic misstep driven more by investor pressure and hype than by clear, user-centric use cases.
Foundations vs. Execution of Apple Intelligence
- Many commenters think Apple built a strong technical and privacy foundation (on‑device models, Private Cloud Compute, hardware NPUs), but shipped weak, fragmented user-facing features.
- Some argue Apple is following its usual pattern: v1 is mediocre but the long-term architecture is sound, so future iterations may become compelling (similar to Maps or early iPhone).
- Others counter that this launch is unusually poor “for Apple”: delayed, underdelivering vs WWDC demos, and often just bad or unfinished.
Privacy, Data, and Trusted Compute
- Debate on whether Apple’s privacy stance conflicts with LLMs:
- One side: LLMs largely train on public data; Apple’s privacy promises mostly concern private user data, so no fundamental clash.
- Other side: even “public” posts (Reddit, papers) shouldn’t automatically be reused for training; Apple’s rhetoric encourages stronger user control.
- Private Cloud Compute is viewed as ambitious and close to a “holy grail” of trusted remote inference. Some are skeptical such guarantees can truly be met; others think it’s the right architecture for regulation and trust.
Feature Quality and UX
- Math Notes is widely discussed:
- Some see it as great—algebraic text/handwriting calculations as an everyday “bicycle for the mind.”
- Others say it doesn’t need LLMs at all; similar functionality existed in tools like Soulver, Calca, Wolfram Alpha, OneNote, etc.
- Notification summaries, mail categorization, and image playground are frequently criticized as inaccurate, uncanny, or “AI slop.” Some enjoy summaries for quick triage and humor.
- Photo “Clean Up” sparks ethical and aesthetic worries about rewriting reality vs traditional editing; others see it as just another post-processing tool.
Comparisons to Competitors and Ecosystem
- Several note that no mobile AI assistant (Apple, Google, Microsoft, Amazon) is reliably good yet; Siri remains weak, but Assistant/Gemini and Alexa are also seen as degraded or ad-heavy.
- Some argue Apple was caught flat-footed by ChatGPT and is now rushing to have an “AI story” for investors; others think they were already deep into ML and mainly rebranding.
- There’s recurring criticism of Apple’s broader software quality (Siri, Maps in many regions, Music, Notes regressions) versus praise for hardware, M‑series chips, and some first‑party apps.
AI’s Broader Value
- Strong split:
- Some assert current LLMs are overblown, a local minimum or mere hype.
- Others report large productivity and domain-specific gains (coding, research, media production, tutoring), arguing that dismissing AI as useless is detached from real-world benefit.