[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-google-deepmind-centralization-gemini-catchup":3,"news-related-bcedeb8e-e5eb-4bbc-98b8-ea12f869055f":41},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":27,"news_slug":34,"published_at":35,"created_at":36,"modified_at":37,"is_published":38,"publish_type":39,"image_url":14,"view_count":40},"bcedeb8e-e5eb-4bbc-98b8-ea12f869055f","Google 收编 DeepMind：25 年最大 AI 重组","Reuters 独家披露 Google 8 月初对 DeepMind 做出 25 年来最大规模人事重组,核心动因是下一代 Gemini 旗舰内部测试显示编程等关键能力仍落后 Anthropic 和 OpenAI,原定 6 月的发布已被推迟约两个月。Brin 4 月内部催促加快研发,Google 通过削弱 DeepMind 自主权、将部分团队并入公司架构,集中力量收归 Gemini 研发。","## Google 又一次站在 AI 竞争的十字路口\n\n据 Reuters 独家报道、TimesNow 转载及 Google 官方博客披露,Google 在 8 月初对其 AI 部门 DeepMind 做出 25 年来最大规模的领导层重组:Demis Hassabis 卸任 CEO,转任新设立的「首席科学家」并改任董事长;资深工程师、原副手 Koray Kavukcuoglu 接棒 CEO;而 Gemini 的两位原始技术 co-lead 加上 Jeff Dean、Oriol Vinyals、Quoc Le 等核心人物离开 Google,成立了一家专注机器学习与科学突破的公益性 AI 企业 Discovery Loop。\n\n让 Alphabet 必须下这一步棋的直接原因并不复杂:Gemini 又被拉开了。\n\n## Gemini 落后,已经不是一两个版本的事\n\nReuters 引述 7 位知情人士的消息指出,Google 原计划 6 月发布的下一代 Gemini 旗舰模型被推迟了约两个月。内部测试结果显示,新模型在编程等关键基准上仍然落后于 Anthropic 和 OpenAI 的同代产品(Solidot《Google 重组 AI 部门旨在赶上竞争对手》)。去年 11 月,Gemini 3.0 曾一度短暂超越竞争对手登顶榜首,但随后 Anthropic 推出 Claude Mythos 5、OpenAI 推出 GPT-5 系列,再次把 Google 推回追赶者位置。\n\n这意味着 Google 已经连续多个版本号处于「发布即落后」的尴尬节奏里:每一次内部评测结果出来,发布日期就被往后推一次。组织层的问题,开始显性地拖累产品节奏。\n\n## Sergey Brin 的「go all in」\n\n矛盾的另一个信号来自公司最高层。据 Reuters\u002FSolidot 报道,Google 联合创始人 Sergey Brin 在今年 4 月的一次 DeepMind 员工大会上罕见现身,公开督促 AI 团队加快研发节奏、集中火力把 Gemini 推上去。这是 Brin 自 2019 年离开日常管理层后,对单一 AI 项目最强烈的一次公开施压。\n\nBrin 亲自下场本身就说明两个问题:第一,Sundar Pichai 主导的 DeepMind 自管理路线没能跑出足够快的速度;第二,董事会层面已经认定 Gemini 是 Google 在生成式 AI 时代「不能输」的战役。\n\n## 把 DeepMind「收回」Google 本部\n\n更深层的组织动作是:DeepMind 正在失去它过去十年的独立性。\n\nReuters 报道指出,8 月 6 日的内部全员会议上,员工被告知部分团队将从 DeepMind 迁出,重新并入 Google 公司架构。这意味着 Google 正在把 DeepMind 从「独立研究机构 + 产品化并行」的混合体,转型为更紧密围绕 Gemini 模型商业化目标的「中央 AI 部门」。\n\n这是一种典型的战略取舍:牺牲研究的长期自由度,换取产品迭代的速度。对 Anthropic 和 OpenAI 这类「以模型为核心」的对手来说,这种收紧反而合理 —— 它们从来都不是研究院形态,而是产品公司形态。Google 现在终于把组织对齐到了同一形状。\n\n## 几个关键看点\n\n**1. Kavukcuoglu 的角色变化是最大信号。** 他从「CTO \u002F 联合创始人」升到「CEO」,意味着 Gemini 的技术负责人第一次拥有了最终拍板权,而不只是向 Hassabis 汇报。这等于把「研究驱动」的旧范式,转向「模型竞争力驱动」的新范式。\n\n**2. Gemini 新一代旗舰已经押后约两个月。** 这个时间窗口对 OpenAI 和 Anthropic 来说,等于一次免费的代际抢跑。如果 Google 在 9-10 月仍无法拿出编程能力对标 Claude Opus 4 \u002F GPT-5.6 的旗舰版本,AI 编程这个细分市场的企业客户预算将被对手进一步锁定。\n\n**3. Discovery Loop 带走的是 Google 最稀缺的研究力量。** Jeff Dean、Oriol Vinyals 这些名字本身就是 Google AI 的「品牌资产」。他们的离开,等同于把 Google Brain 时代的最后一批研究型灵魂送进了公益性机构。对学术界是个好事,但对 Google 商业 AI 是个明显的智力外流。\n\n## 所以呢?\n\nGoogle 这一轮重组,本质上是在承认一个事实:单靠「研究自治 + 巨大算力 + 品牌优势」这个组合,已经打不过 Anthropic 和 OpenAI 的「产品化工程化 + 闭源旗舰 + 垂直客户」这套打法。\n\n把 DeepMind 收归中央、Kavukcuoglu 接棒、Brin 亲自催进度 —— 这一套组合拳之后,市场要看的不是发布会,而是 2026 年 Q4 之前,Google 能不能拿出一款在编程、企业 Agent、长上下文三个核心场景里都站得住的下一代 Gemini。如果拿不出,这轮重组只会把 Google 推向「组织换帅 → 仍然落后 → 继续重组」的循环。\n\nAnthropic Mythos、OpenAI GPT-5.6 已经把发令枪打响。Google 现在是唯一一个同时需要追赶模型能力、追赶商业化、追赶组织效率的三线选手。25 年来最大的 AI 重组,听起来很硬核,但真正硬的仗还在后面。\n\n---\n\n**参考来源**:Reuters 独家报道(Via TimesNow 转载,https:\u002F\u002Fwww.timesnownews.com\u002Ftechnology-science\u002Fgoogle-ai-shake-up-explained-gemini-struggling-to-keep-up-with-openai-and-anthropic-article-155604258);Solidot《Google 重组 AI 部门旨在赶上竞争对手》(https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85082);Google 官方 CEO 公开信。","https:\u002F\u002Fwww.timesnownews.com\u002Ftechnology-science\u002Fgoogle-ai-shake-up-explained-gemini-struggling-to-keep-up-with-openai-and-anthropic-article-155604258","ea95d933-6860-4081-9970-cede7c107cd6",[11,15,18,21,24],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"a9524a82-a7c5-4daa-bb4b-a7ee77bb0b94","gemini",{"id":19,"name":20,"slug":20,"description":14,"color":14},"8cf7490f-2449-4ba7-be19-61befa0d92b4","google",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":25,"name":26,"slug":26,"description":14,"color":14},"95d7995a-fddb-47ba-b8e6-e976ac65414b","strategy",[28],{"id":29,"lang":30,"title":31,"summary":32,"content":33},"660c7d81-334e-4125-be78-030d19c0e060","en","Google folds DeepMind in: its biggest AI reshuffle in 25 years","Reuters reveals Google made its largest DeepMind leadership reshuffle in 25 years in early August. The core driver: internal tests show the next-generation Gemini flagship still trails Anthropic and OpenAI in coding and other key capabilities, and the original June launch has been pushed back roughly two months. Brin pressed internally in April for faster progress. Google is also reducing DeepMind's autonomy and folding selected teams into the broader company structure, concentrating firepower on Gemini development.","## Google Stands at the AI Crossroads Once Again\n\nAccording to a Reuters exclusive (republished by TimesNow) and Google's official blog, Google carried out the largest leadership reshuffle of its AI division DeepMind in 25 years in early August. Demis Hassabis stepped down as CEO, taking on a newly created Chief Scientist role and becoming chairman. Senior engineer and former deputy Koray Kavukcuoglu took over as CEO. Meanwhile, the two original technical co-leads of Gemini, plus Jeff Dean, Oriol Vinyals, Quoc Le and other core figures, left Google to found a public-benefit AI company called Discovery Loop, focused on machine learning, science, and engineering breakthroughs.\n\nThe direct reason Alphabet had to make this move is straightforward: Gemini has fallen behind again.\n\n## Gemini's Lag Is No Longer a One-Version Problem\n\nCiting seven people familiar with the matter, Reuters reported that Google's next flagship Gemini model, originally slated for June, has been delayed by roughly two months. Internal testing showed the new model still trails Anthropic and OpenAI's same-generation products on key benchmarks including coding (as covered in Solidot's \"Google Restructures AI Division to Catch Competitors\"). Last November, Gemini 3.0 briefly overtook competitors to top the leaderboards, but Anthropic's Claude Mythos 5 and OpenAI's GPT-5 line soon pushed Google back into a catch-up position.\n\nThis means Google has spent multiple consecutive version numbers in an awkward rhythm of \"ship, then immediately fall behind\": every internal benchmark review pushes the launch date further out. Organizational issues are now visibly dragging product cadence.\n\n## Sergey Brin's \"Go All In\"\n\nAnother signal of the tension came from the very top. Per Reuters and Solidot, Google co-founder Sergey Brin made a rare appearance at a DeepMind all-hands in April this year, publicly urging the AI team to speed up and concentrate firepower on pushing Gemini forward. It was the strongest public pressure Brin has applied to a single AI project since stepping back from day-to-day management in 2019.\n\nBrin's personal involvement signals two things. First, the DeepMind self-management route under Sundar Pichai has not produced speed fast enough. Second, the board-level consensus is that Gemini is the one battle Google cannot afford to lose in the generative-AI era.\n\n## Folding DeepMind Back Into the Center\n\nThe deeper organizational move: DeepMind is losing the independence it has had for the past decade.\n\nPer Reuters, at the August 6 internal all-hands, employees were told that some teams would move out of DeepMind and back into the broader Google corporate structure. This means Google is converting DeepMind from a hybrid of \"independent research lab plus parallel productization\" into a tighter, centrally-run AI division focused on the commercial goals of the Gemini model.\n\nIt is a classic strategic trade-off: sacrifice long-term research latitude in exchange for product iteration speed. For Anthropic and OpenAI, which are fundamentally model-centric opponents rather than research-institute-shaped ones, this tightening actually makes sense. Google is finally aligning its organization to the same shape as its rivals.\n\n## Three Things to Watch\n\n**1. Kavukcuoglu's role change is the biggest signal.** His move from CTO\u002Fco-founder to CEO means the technical lead of Gemini now has final say, not just a reporting line to Hassabis. The shift effectively rotates the old \"research-driven\" paradigm toward a \"model-competitiveness-driven\" one.\n\n**2. The next Gemini flagship is already pushed back about two months.** For OpenAI and Anthropic, that window is a free generational head start. If Google cannot ship, by September or October, a flagship whose coding capability matches Claude Opus 4 \u002F GPT-5.6, enterprise budgets in the AI coding segment will lock in further with the competitors.\n\n**3. Discovery Loop takes with it some of Google's most scarce research talent.** Names like Jeff Dean and Oriol Vinyals are themselves Google AI's brand assets. Their departure ships the last research-driven souls of the Google Brain era into a public-benefit organization. Good for academia, but a visible intellectual outflow from Google's commercial AI.\n\n## So What?\n\nThis restructure is, in essence, an admission that the combination of \"research autonomy + massive compute + brand\" is no longer enough to beat Anthropic and OpenAI's playbook of \"product-grade engineering + closed flagship + vertical customers\".\n\nWith DeepMind centralized, Kavukcuoglu at the helm, and Brin personally pushing for speed, the market is not watching for the next launch event. It is watching whether Google can ship, before Q4 2026, a next-generation Gemini that holds up in three core scenarios: coding, enterprise agents, and long context. If not, this round of reshuffling will only push Google into the loop of \"change leadership → still fall behind → reshuffle again\".\n\nAnthropic Mythos and OpenAI GPT-5.6 have already fired the starting gun. Google is the only player that now has to chase on three fronts at once: model capability, commercialization, and organizational efficiency. The largest AI reshuffle in 25 years sounds dramatic — but the real fight is still ahead.\n\n---\n\n**Sources**: Reuters exclusive report (via TimesNow republication, https:\u002F\u002Fwww.timesnownews.com\u002Ftechnology-science\u002Fgoogle-ai-shake-up-explained-gemini-struggling-to-keep-up-with-openai-and-anthropic-article-155604258); Solidot, \"Google Restructures AI Division to Catch Competitors\" (https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85082); Google official CEO letter.","google-deepmind-centralization-gemini-catchup","2026-08-14T08:00:00Z","2026-08-14T05:05:52.142355Z","2026-08-19T01:48:03.231362Z",true,"agent",228,{"items":42},[43,48,53,58,63,68],{"id":44,"title":45,"news_slug":46,"published_at":47},"4bd8e8bd-7066-4ab7-bd97-e24ea3921395","Gemini 因编程落后推迟两月:Brin 4 月督促背后,Google 把研发「收回到一个人」手里的组织账本","google-gemini-coding-behind-deepmind-reshuffle","2026-08-14T03:30:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"df25ef84-5f0b-477b-9940-a0da477d8169","DeepMind 新主帅接棒：Hassabis 退任，Gemini 4 成 Google 筹码","deepmind-kavukcuoglu-gemini-4-reshuffle","2026-08-07T00:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"7b9cdf6e-5ef0-4ece-ab6c-e8cec1b02397","Google 重组 DeepMind 领导层,Gemini 研发提速应对 Anthropic 与 OpenAI 竞争","google-deepmind-reshuffle-gemini-speed","2026-08-25T07:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"8b6c20ec-7222-48cf-af2c-ac97466a2b0a","Gemini 月活破 10 亿:Google 第一次把 AI 助手做成自家「最快十亿用户产品」","gemini-app-1b-monthly-users","2026-08-12T03:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"34edaffc-6b5c-4df1-9e2f-d864cada6063","Gemini 走进 K-12 课堂：Google 把「上下文」塞进每个作业","gemini-classroom-k12-contextualized-prompts","2026-08-07T02:00:00+00:00",{"id":69,"title":70,"news_slug":71,"published_at":72},"0a4aa266-a1f0-4306-accc-47fe556db43f","Google 25 年最大 AI 重组：Hassabis 卸任 DeepMind CEO","google-deepmind-shakeup-hassabis-dean-discovery-loop-0806","2026-08-06T00:00:00+00:00"]