文章强调路线之争与资源规模：ChatGPT 以大型语言模型引爆热潮，背后依赖价值「数千亿美元」等级的算力投入；哈萨比斯则押注「世界模型」，用模拟与物理世界训练 AI，而非只靠聊天机器人。竞争面上，ChatGPT 在用户数与企业化推进上领先，且有约两年的企业市场先发；Google 的 Gemini Enterprise 直到 10 月才推出。
数据对比呈现追赶趋势：Gemini 3 登顶多个效能榜单，并让 OpenAI 内部出现「code red」警报；其月活 app 用户超过 6.5 亿，透过搜寻 AI Overviews 触及估计约 20 亿用户。Google 于 2024 年 8 月以约 27 亿美元重新网罗 Transformer 共同发明者 Noam Shazeer（他曾于 2021 年离职），外界传他修复关键效率问题。压力在于变现：AlphaFold 促进约 300 万名科学家研究，但迄今仍未直接产出 FDA 核准药物；若「懂物理」的眼镜能成真并热卖，可能成为 AI 的杀手级应用与哈萨比斯的商业转折。
Since Google acquired DeepMind in 2014, Demis Hassabis has delivered major scientific prestige (from landmark game victories to prize-winning protein-folding work), but a true product breakthrough has been elusive. In 2023 Alphabet consolidated its two AI groups under his leadership. Google now aims to launch AI smart glasses in 2026 with Samsung, including an in-lens display for navigation and translation—an attempt to redeem the smart-glasses category and reverse the setback sparked by ChatGPT three years earlier.
The article frames a strategic split: Chatbots and large language models drove the boom, backed by computing power valued at “hundreds of billions of dollars,” while Hassabis argues that simulation-trained “world models” will unlock the next leap. Competitive metrics still favor OpenAI: ChatGPT leads in user scale and has roughly a two-year head start in enterprise rollout, while Gemini Enterprise only arrived in October. Still, Google’s pace is accelerating.
Gemini 3 topped performance leaderboards, triggered a reported “code red” at OpenAI, and reached over 650 million monthly app users, plus an estimated 2 billion users via AI Overviews in Search. Google also effectively spent $2.7 billion in August 2024 to bring back Noam Shazeer (who left in 2021), credited with key efficiency gains. Hassabis’ bet is that physics-aware assistants (e.g., Project Astra) embedded in glasses could become AI’s killer app, as monetization pressure grows: AlphaFold is used by about 3 million scientists yet has produced no FDA-approved drugs so far.
