Apple Has Failed At Artificial Intelligence. Again.

Summary — “Apple Has Failed At Artificial Intelligence. Again.”

Main claim Apple abandoned building its own large language models (LLMs) and entered a multi‑year deal to base its “Apple Foundation models” on Google’s Gemini in the cloud. The narrator frames this as a failure after roughly two years of trying to compete with ChatGPT‑style services.

Key technological and product points

Siri’s foundational problems

• Largely rule‑based architecture (legacy approach from ~2011), causing brittle natural‑language handling. Example: failing to answer simple time‑zone questions unless phrased exactly.
• Narrow scope (stuck at ~15 domains). Apple delayed and tightly controlled third‑party extensions for years, unlike Alexa which opened quickly and scaled.

Privacy and on‑device focus

• Apple’s insistence on encrypted, on‑device processing limits its ability to train large cloud models and to leverage broad personal data — a tradeoff that hampers modern LLM approaches.

Secrecy and lack of academic openness

• Internal secrecy (multiple AI teams unaware of each other and a refusal to publish research) slowed progress and made recruiting/retention of top AI researchers difficult.

Underinvestment and talent issues

• Reportedly far fewer AI training chips (≈100k) versus Microsoft (≈500k) and Google (>1M).
• Apple didn’t match pay packages of competitors, contributing to recruitment/retention problems and high turnover (example cited: 12 departures in one quarter).

Product and marketing choices

• Apple’s product culture favors curated, closed experiences (“we’ll do it for you”) rather than public chatbots; this conflicted with the paradigms that made ChatGPT successful.
• “Apple Intelligence” branding was vague; executives (John Giannandrea and Craig Federighi) appeared uncertain in interviews about what qualifies as Apple Intelligence or whether those features used Apple LLMs.

Early Apple AI rollout (iOS 18 / 2024) and quality issues

• Several beta features were criticized or failed in practice:
  • Email summaries prioritized scam emails.
  • Generative features (referred to as “Genoji” in the transcript) did not match demos and drained battery.
  • Notification summaries misrepresented headlines (feature was later disabled).
• A promised rebuild of Siri was delayed; internal chatbot “Ajax” was reportedly ~25% less accurate than ChatGPT.

Business and strategic consequences

• Apple’s stock underperformed peers after ChatGPT’s rise, putting pressure on leadership to announce modern AI; the Apple Intelligence reveal boosted the stock (~7% the day after, ~30% by year end).
• Relying on Google’s Gemini creates long‑term strategic risk (historical precedent: Apple’s dependence on Google Maps before building its own maps).
• Rumors indicate Apple will initially build new features on Gemini (improved Siri, possible Siri chatbot by 2027); moving to proprietary models will likely take years.

Criticized or impacted features (from the video)

• Siri (voice assistant) — core rebuild delayed; brittle NLP
• Type to Siri — accessibility feature included under Apple Intelligence
• Email summaries — prioritized scams, poor performance
• Notification summaries for news — misrepresented headlines; temporarily disabled
• Generative features (demoed) — didn’t match ads and consumed significant battery
• Ajax (internal chatbot) — less accurate than ChatGPT

Analysis / Diagnosis

The failure is attributed to a “perfect storm” of Apple weaknesses:

• Legacy rule‑based technology and narrow domain focus
• Desire for tight control and curated experiences
• Extreme privacy constraints limiting cloud training and data access
• Internal secrecy and refusal to publish research
• Conservative spending and product planning, insufficient hardware scale
• Shareholder and market pressures that forced a faster, risk‑averse response

Together, these factors slowed Apple’s ability to build competitive LLMs at the scale and pace of Google, OpenAI, or Microsoft.

Follow‑ups mentioned

• The creator will publish a deeper video on the original Siri failure history.

Main speakers / sources cited

• Video host: Greg (from “Apple Explained”)
• Apple executives featured: John Giannandrea, Craig Federighi
• Interviewer referenced: John Gruber
• Historical figures/context: Steve Jobs, Tim Cook, Scott Forstall (Maps context)
• External companies/technologies referenced: Google (Gemini), OpenAI (ChatGPT), Microsoft, Amazon (Alexa), Meta, Anthropic, Tencent, IBM, Oracle
• Internal Apple projects referenced: Ajax (internal chatbot), “Apple Intelligence” (branding), iOS 18 AI features