Google is winning on every AI front

Original Article ↗ Hacker News Discussion ↗

Trust, privacy & who uses Google Cloud

  • Some argue Google has lost brand trust, especially with governments and large enterprises; others counter that many big firms (banks, pension funds, F500) are on GCP, at least for ML.
  • Several current/former insiders claim Google’s internal security and data-handling controls are exceptionally strong.
  • A recurring distinction:
    • Data Google collects about you for ads is seen as overreaching.
    • Data you store in Drive/Cloud is seen as among the safest consumer options.
  • Many still avoid Google on principle (“I don’t want to feed the ad machine”), even if they acknowledge Gemini’s quality.

Gemini vs ChatGPT/Claude/Grok: UX and capability

  • Gemini 2.5 Pro is widely praised: fast, cheap (often free), huge context window, strong coding and “Deep Research,” especially via AI Studio.
  • But the consumer Gemini app and mobile assistant are heavily criticized:
    • Worse than the old Google Assistant for alarms, timers, calling, weather; confusing settings; localization issues.
    • Guardrails and refusals trigger on mundane or mildly controversial topics, making it feel “sterile” vs ChatGPT or Grok.
  • For coding, opinions split: many find Gemini 2.5 top-tier (especially with tools like Roo Code, Aider, etc.), others say Claude 3.7 or GPT‑4o still produce more reliable code and better tool-calling.

Search, AI Overviews & business model tension

  • Google’s AI Overviews and search integration are seen as one of its weakest AI fronts: low factual quality, odd answers, and degraded overall search UX.
  • Several commenters frame Google’s dilemma:
    • LLMs plus retrieval can obsolete traditional ranking advantages.
    • Moving too fast cannibalizes ad-heavy search; moving too slow cedes mindshare to OpenAI/Perplexity.
  • Many expect chatbots to go ad-supported; there is intense concern about “stealth” ads and behavioral manipulation embedded in conversational output.

TPUs, infrastructure & “moats”

  • TPU + JAX stack is highlighted as a major structural advantage: vertical integration, perf-per-watt, freedom from Nvidia, and huge internal TPU clusters, especially for ads.
  • Others argue TPUs haven’t yielded a clear external lead, were often over-specialized, and that Nvidia’s general-purpose CUDA ecosystem has been more agile.
  • Consensus: training/inference at frontier scale will favor a few players with massive compute, data, and distribution; Google is one of them, but not unassailable.

Market dynamics: winner-takes-most or commodity?

  • Several see models converging in quality and becoming semi‑commodities; differentiation may shift to:
    • Integration with ecosystems (Android, Workspace, YouTube, Office, etc.)
    • UX, agent frameworks, and trust.
  • Low switching costs (chat-style APIs, multiple frontends) mean no one has durable lock-in yet; many power users hop between Gemini, Claude, GPT, Grok, DeepSeek.
  • Ex‑OpenAI voices suggest OpenAI’s top research talent has thinned and that subscription revenue may cap out, while Google can subsidize AI through ads and bundles—but others note OpenAI still leads in brand, consumer mindshare, and some modalities (voice, images).

Guardrails, localization & “vibes”

  • Gemini is viewed as over-censored and U.S.-centric in some locales (week start, units, language variants, sensitive topics), hurting its suitability as a general assistant.
  • ChatGPT is often preferred for “friend-like” conversations and memory, Claude for tone and creative writing, Grok for less-filtered answers.
  • Many agree: technically, Google is now highly competitive or ahead on several benchmarks and infra; on trust, product polish, and cultural “vibes,” it is far from “winning on every AI front.”