• 榜单数据10-19 11:33
  • 论文数据10-16 13:30
  • 周报10-18 20:50

Gemini(Google DeepMind) · 引语索引

机构深度专题更新于 AI 起草 · 人工审核已核查 · 已审核审核人 负责人

从编年史、人物与路线差异到基础设施、Agent 与 AI for Science;事实截止日 2026-07-25(原型口径)。内网直播综述材料已剔除。

共 0 条引语

引语主题出处与定位来源等级原话状态

证伪区:流传但无法证实或已被证伪的说法

每条都写明依据。这些说法不是原话引用,正文也不以它们为依据。

  • 无法证实I've always believed that it's gonna be the most important invention that humanity will ever make.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.cbsnews.com/news/google-artificial-intelligence-future-60-minutes-transcript-2023-04-16/(在新窗口打开)2023-04-16CBS《60 Minutes》"The Revolution"(记者 Scott Pelley);Demis Hassabis权威媒体非本人原话

  • 无法证实My Proposed Conjecture: "Any pattern that can be generated or found in nature can be efficiently discovered and modelled by a classical learning algorithm"

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.nobelprize.org/uploads/2024/12/hassabis-lecture.pdf(在新窗口打开)2024-12-08诺贝尔奖演讲讲稿幻灯片(他本人为这句话自加的标题即「My Proposed Conjecture」);Demis Hassabis权威媒体非本人原话

  • 无法证实Maybe I'll just make some money... and just work on AI research for a long time.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/jeff-dean-and-noam-shazeer(在新窗口打开)2025-02-12Dwarkesh Podcast,回忆当初为什么加入 Google;Noam Shazeer权威媒体非本人原话

  • 无法证实I've always had this sort of screaming at me in my face and it's getting louder.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://lexfridman.com/demis-hassabis-2-transcript/(在新窗口打开)2025-07-23Lex Fridman Podcast #475,谈自己做 AI 的最终动机;Demis Hassabis权威媒体非本人原话

  • 无法证实while my mode is about 2025, my expected value is actually a bit higher at 2028… my 90% credibility region… comes out at 2018 to 2036.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.vetta.org/2009/12/tick-tock-tick-tock-bing/(在新窗口打开)2009-12-07个人博客 vetta.org《Tick, tock, tick, tock… BING》;Shane Legg权威媒体非本人原话

  • 无法证实I will not hazard anything other than a broad guess, which is I think we are fifteen to fifty years away—probably closer to the lower end of that range.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.forbes.com/sites/peterhigh/2016/08/01/a-tour-inside-googles-brain/(在新窗口打开)2016-08-01Metis Strategy 播客(2016-07-18 录制),Forbes 发表编辑版问答,Peter High 问 AGI 还有多远;Jeff Dean权威媒体非本人原话

  • 无法证实I think there's a 50% chance that we have AGI by 2028. Now, it's just a 50% chance.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/shane-legg(在新窗口打开)2023-10-26Dwarkesh Podcast(时间戳 00:34:03 起);Shane Legg权威媒体非本人原话

  • 无法证实I think it's entirely plausible but I'm not going to be surprised if it doesn't happen by then.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/shane-legg(在新窗口打开)2023-10-26Dwarkesh Podcast,紧接上一条;Shane Legg权威媒体非本人原话

  • 无法证实a machine that can do the sorts of cognitive things that people can typically do, possibly more

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/shane-legg(在新窗口打开)2023-10-26Dwarkesh Podcast,给 AGI 下定义时;Shane Legg权威媒体非本人原话

  • 无法证实we stated in there [a] 20-year time scale, and I think we're actually pretty on track.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=nwUARJeeplA(在新窗口打开)2024-02-23NYT《Hard Fork》播客;Demis Hassabis权威媒体非本人原话

  • 无法证实I'm not sure I agree completely, but it's a pretty interesting thought experiment.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/jeff-dean-and-noam-shazeer(在新窗口打开)2025-02-12Dwarkesh Podcast(回应 Noam Shazeer「世界 GDP 会涨几个数量级」);Jeff Dean权威媒体非本人原话

  • 无法证实In the next five to ten years, I think.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.cbsnews.com/news/artificial-intelligence-google-deepmind-ceo-demis-hassabis-60-minutes-transcript/(在新窗口打开)2025-04-20CBS《60 Minutes》,记者 Scott Pelley 问「On track for AGI in—」;Demis Hassabis权威媒体非本人原话

  • 无法证实Will the AI think it has reached AGI by 2030? I would say we will just fall short of that timeline, so I think it'll take a bit longer.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://lexfridman.com/sundar-pichai-transcript/(在新窗口打开)2025-06-05Lex Fridman Podcast #471(时间戳 00:40:31–00:41:24);Sundar Pichai权威媒体非本人原话

  • 无法证实I don't think we'll quite get there by 2030, so my sense is it's slightly after that.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://lexfridman.com/sundar-pichai-transcript/(在新窗口打开)2025-06-05Lex Fridman Podcast #471,紧接上一条;Sundar Pichai权威媒体非本人原话

  • 无法证实I don't know who used it first, maybe Karpathy did.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://lexfridman.com/sundar-pichai-transcript/(在新窗口打开)2025-06-05Lex Fridman Podcast #471,谈 AJI(artificial jagged intelligence)时;Sundar Pichai权威媒体非本人原话

  • 无法证实My estimate is sort of 50% chance by in the next five years, so by 2030 let's say.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://lexfridman.com/demis-hassabis-2-transcript/(在新窗口打开)2025-07-23Lex Fridman Podcast #475(时间戳 00:52:33);Demis Hassabis权威媒体非本人原话

  • 无法证实The reason I tend to steer away from AGI conversations is lots of people have very different definitions of it and the difficulty of the problem varies by like factors of a trillion.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=OEuh89BWRL4(在新窗口打开)2025-08-21The Moonshot Podcast: Deep Dive;Jeff Dean权威媒体非本人原话

  • 无法证实No, I think it's dead on track if you see what I mean… I think this is the fastest progress that anyone in the industry has managed.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=rq-2i1blAlU(在新窗口打开)2025-11-18NYT《Hard Fork》播客(Gemini 3 发布同日);Demis Hassabis权威媒体非本人原话

  • 无法证实Today we're taking another big step on the path toward AGI and releasing Gemini 3.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://blog.google/products-and-platforms/products/gemini/gemini-3/(在新窗口打开)2025-11-18Gemini 3 发布官方博客(Demis Hassabis 与 Koray Kavukcuoglu 署名);Google(官方,Gemini 3 发布)权威媒体非本人原话

  • 无法证实AI systems building AI systems.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=02YLwsCKUww(在新窗口打开)2026-01-20WEF 达沃斯主论坛《The Day After AGI》(主持人:The Economist 主编 Zanny Minton Beddoes);Dario Amodei权威媒体非本人原话

  • 无法证实I agree on that.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=02YLwsCKUww(在新窗口打开)2026-01-20WEF 达沃斯主论坛《The Day After AGI》(回应 Amodei);Demis Hassabis权威媒体非本人原话

  • 无法证实When we look back at this time, I think — maybe I'm thinking 10 years from now — I think we will realize that we were standing in the foothills of the singularity now.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=QhdEJFFaig0(在新窗口打开)2026-05-19(原话)/2026-06 复述Google I/O 2026 keynote 收尾;**全文以他本人在 Stanford GSB 上的复述为准**("So **the full thing I said** to sort of close the conference with was…")[@src-1754eee058]他本人复述的全文 ·;Demis Hassabis权威媒体非本人原话

  • 无法证实the compression objective is the only training objective that I know right now ... which is completely non-gameable

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=dO4TPJkeaaU(在新窗口打开)2023-02Stanford 演讲《Compression for AGI》;Jack Rae权威媒体非本人原话

  • 无法证实You set off AlphaZero in the morning and it starts off playing randomly. By lunchtime it's able to beat me… and then by the evening, it's stronger than the world champion.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.cbsnews.com/news/google-artificial-intelligence-future-60-minutes-transcript-2023-04-16/(在新窗口打开)2023-04-16CBS《60 Minutes》"The Revolution";Demis Hassabis权威媒体非本人原话

  • 无法证实a series of powerful, multimodal AI models

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://blog.google/innovation-and-ai/technology/ai/april-ai-update/(在新窗口打开)2023-04-20合并备忘录(描述 Jeff Dean 的 Chief Scientist 新角色与第一个战略项目);Sundar Pichai权威媒体非本人原话

  • 无法证实At a high level you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.wired.com/story/google-deepmind-demis-hassabis-chatgpt/(在新窗口打开)2023-06-26WIRED 专访;Demis Hassabis权威媒体非本人原话

  • 无法证实I don't see a wall.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/shane-legg(在新窗口打开)2023-10-26Dwarkesh Podcast;Shane Legg权威媒体非本人原话

  • 无法证实The Gemini models are natively multimodal, as they are trained jointly across text, image, audio, and video.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://arxiv.org/abs/2312.11805(在新窗口打开)2023-12arXiv:2312.11805 §4;Gemini 1.0 技术报告(官方)权威媒体非本人原话

  • 无法证实the models are multimodal from the beginning and can natively output images using discrete image tokens

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://arxiv.org/abs/2312.11805(在新窗口打开)2023-12arXiv:2312.11805 §2,说明与 Flamingo / CoCa / PaLI 的区别;Gemini 1.0 技术报告(官方)权威媒体非本人原话

  • 无法证实Until now, the standard approach… involved training separate components for different modalities and then stitching them together… We designed Gemini to be natively multimodal, pre-trained from the start on different modalities.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://blog.google/innovation-and-ai/technology/ai/google-gemini-ai/(在新窗口打开)2023-12-06Gemini 1.0 发布官方博客,「A note from Google and Alphabet CEO Sundar Pichai」段落内;Sundar Pichai权威媒体非本人原话

  • 无法证实running up to 1 million tokens consistently, achieving the longest context window of any large-scale foundation model yet

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://blog.google/innovation-and-ai/products/google-gemini-next-generation-model-february-2024/(在新窗口打开)2024-02-15Gemini 1.5 发布官方博客;Demis Hassabis权威媒体非本人原话

  • 无法证实Gemini 1.5 Pro also shows impressive 'in-context learning' skills, meaning that it can learn a new skill from information given in a long prompt, without needing additional fine-tuning.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://blog.google/innovation-and-ai/products/google-gemini-next-generation-model-february-2024/(在新窗口打开)2024-02-15Gemini 1.5 发布官方博客;Google DeepMind(官方)权威媒体非本人原话

  • 无法证实You can think of [long context] as the model's working memory… and the precision of recalling things from that long context is also quite important.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=nwUARJeeplA(在新窗口打开)2024-02-23NYT《Hard Fork》播客;Demis Hassabis权威媒体非本人原话

  • 无法证实AlphaZero-like planning mechanisms on top… this is what's missing from our current large models.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/demis-hassabis(在新窗口打开)2024-02-28Dwarkesh Podcast;Demis Hassabis权威媒体非本人原话

  • 无法证实probably about the maximum

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/demis-hassabis(在新窗口打开)2024-02-28Dwarkesh Podcast,谈代际之间的规模跨度;Demis Hassabis权威媒体非本人原话

  • 无法证实Our history in agents is longer than our generalized model work.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.theverge.com/2024/5/14/24156296/google-ai-gemini-astra-assistant-live-io(在新窗口打开)2024-05-14The Verge 专访(Google I/O 2024,Project Astra 发布日);Demis Hassabis权威媒体非本人原话

  • 无法证实Arithmetic is very, very cheap, and moving data around is comparatively much more expensive.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/jeff-dean-and-noam-shazeer(在新窗口打开)2025-02-12Dwarkesh Podcast;Noam Shazeer权威媒体非本人原话

  • 无法证实As we know from our experience with AlphaGo, responses improve when we give these models more time to think.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://singjupost.com/transcript-of-google-ceo-sundar-pichai-at-google-i-o-2025-keynote/(在新窗口打开)2025-05-20Google I/O 2025 keynote(发布 Gemini 2.5 Pro Deep Think);Demis Hassabis权威媒体非本人原话

  • 无法证实Post-training: Still a total greenfield.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://threadreaderapp.com/thread/1990854455802343680.html(在新窗口打开)2025-11X 平台发帖(Gemini 3 发布前后);Oriol Vinyals权威媒体非本人原话

  • 无法证实We work quite hard on the persona, which we call it internally… more succinct, more to the point, helpful… I find it more pleasant to brainstorm with.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=rq-2i1blAlU(在新窗口打开)2025-11-18NYT《Hard Fork》播客(谈 Gemini 3 的人格设计);Demis Hassabis权威媒体非本人原话

  • 无法证实You paid a thousand picojoules in order to do your one picojoule multiply.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.latent.space/p/jeffdean(在新窗口打开)2026-02-12Latent Space 播客;Jeff Dean权威媒体非本人原话

  • 无法证实you're trying to design a chip starting today and that design might take two years before it even lands in a data center… So you're trying to predict two to six years out what ML computations will people want to run… having people with interesting ML research ideas really enables us to then get interesting hardware features put into TPU N plus two.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.latent.space/p/jeffdean(在新窗口打开)2026-02-12Latent Space 播客,谈软硬协同;Jeff Dean权威媒体非本人原话

  • 无法证实I agree with you, we're kind of using duct tape right now. So like shove it all in the context window.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.ycombinator.com/library/P3-how-to-build-the-future-demis-hassabis(在新窗口打开)2026-04-30Y Combinator《How to Build the Future》访谈,回应 Garry Tan;Demis Hassabis权威媒体非本人原话

  • 无法证实a million tokens isn't that much. It's only like 20 minutes.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.ycombinator.com/library/P3-how-to-build-the-future-demis-hassabis(在新窗口打开)2026-04-30Y Combinator《How to Build the Future》访谈;Demis Hassabis权威媒体非本人原话

  • 无法证实AlphaGo was an agent. Even our original Atari work… they were agents. Maybe we were a bit ahead of our time.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.fastcompany.com/91544235/demis-hassabis-google-io-2026(在新窗口打开)2026-05Fast Company 专访(I/O 2026 期间,谈 Gemini Spark);Demis Hassabis权威媒体非本人原话

  • 无法证实Last year, I outlined our vision of extending Gemini's incredible multimodal capabilities to become a world model… Now we're taking the next big step. I'm excited to announce Gemini Omni. Our new model that can create anything from any input.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=QhdEJFFaig0(在新窗口打开)2026-05-19Google I/O 2026 keynote(发布 Gemini Omni);Demis Hassabis权威媒体非本人原话

  • 无法证实We're starting with video, but over time, Omni will be able to generate any output from any input. That's always been our goal for Gemini, and it's why we built it multimodal from the start.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=QhdEJFFaig0(在新窗口打开)2026-05-19Google I/O 2026 keynote(Gemini Omni);Demis Hassabis权威媒体非本人原话

  • 无法证实I felt like we would make more progress on continual learning and more not-so-structured model architectures… we are not doing that yet. But what we're doing seems to work.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=8hfpLa5wPGo(在新窗口打开)2026-05-29Gemini co-leads 官方圆桌;Jeff Dean权威媒体非本人原话

  • 无法证实To disagree a bit though, we're pre-trained. I mean, it's not like you're the first human.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=8hfpLa5wPGo(在新窗口打开)2026-05-29Gemini co-leads 圆桌(当场反驳 Jeff Dean 的「人类样本效率高一千倍」);Oriol Vinyals权威媒体非本人原话

  • 无法证实It feels almost impossible, if you asked me a few years ago, this approach would work… But it did happen.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=8hfpLa5wPGo(在新窗口打开)2026-05-29Gemini co-leads 官方圆桌,谈世界模型能力涌现;Oriol Vinyals权威媒体非本人原话

  • 无法证实We have philosophers here and ethicists, but also other types of scientists, physicists… I tried to build a sort of new type of Bell Labs, but in its golden era.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://podscripts.co/podcasts/lex-fridman-podcast/299-demis-hassabis-deepmind(在新窗口打开)2022-07-01Lex Fridman Podcast #299;Demis Hassabis权威媒体非本人原话

  • 无法证实There's just too much brand risk in large companies to ever launch anything fun.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://a16z.com/universally-accessible-intelligence/(在新窗口打开)2023a16z 访谈《Universally accessible intelligence》;Noam Shazeer权威媒体非本人原话

  • 无法证实By creating Google DeepMind, I believe we can get to that future faster.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://deepmind.google/blog/announcing-google-deepmind/(在新窗口打开)2023-04-20《Announcing Google DeepMind》内部信 / 官方博客;Demis Hassabis权威媒体非本人原话

  • 无法证实we need to work with greater speed, stronger collaboration and execution, and to simplify the way we make decisions to focus on achieving the biggest impact

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://deepmind.google/blog/announcing-google-deepmind/(在新窗口打开)2023-04-20《Announcing Google DeepMind》内部信 / 官方博客;Demis Hassabis权威媒体非本人原话

  • 无法证实The research advances from the phenomenal Brain and DeepMind teams laid much of the foundations of the current AI industry, from Deep Reinforcement Learning to Transformers.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://deepmind.google/blog/announcing-google-deepmind/(在新窗口打开)2023-04-20《Announcing Google DeepMind》内部信 / 官方博客;Demis Hassabis权威媒体非本人原话

  • 无法证实80 or 90 percent of the innovations come from one or the other.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.wired.com/story/google-deepmind-demis-hassabis-chatgpt/(在新窗口打开)2023-06WIRED 专访(谈 Brain 与 DeepMind 合并的理由);Demis Hassabis权威媒体非本人原话

  • 无法证实Eventually we pulled DeepMind into Google and in a way DeepMind was never independent and isn't independent now.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=lSShMt11y_w(在新窗口打开)2023-0980,000 Hours 播客;Mustafa Suleyman权威媒体非本人原话

  • 无法证实Google has spent the last 15 years in Europe, on the witness stand… if Google then goes and creates an experience that disintermediates the third party content creator, then that undermines everything.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.fastcompany.com/90959853/mustafa-suleyman-inflection-pi(在新窗口打开)2023-09-29Fast Company 专访;Mustafa Suleyman权威媒体非本人原话

  • 无法证实the first realization of the vision we had when we formed Google DeepMind earlier this year

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://blog.google/innovation-and-ai/technology/ai/google-gemini-ai/(在新窗口打开)2023-12-06Gemini 1.0 发布官方博客,「A note from Google and Alphabet CEO Sundar Pichai」段落内;Sundar Pichai权威媒体非本人原话

  • 无法证实All the user prompts and outputs in the video are real, shortened for brevity.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.theverge.com/2023/12/7/23992737/google-gemini-misrepresentation-ai-accusation(在新窗口打开)2023-12-07X 帖(回应 Gemini 1.0 "Hands-on" 演示视频争议);Oriol Vinyals权威媒体非本人原话

  • 无法证实We made it to inspire developers.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.theverge.com/2023/12/7/23992737/google-gemini-misrepresentation-ai-accusation(在新窗口打开)2023-12-07X 帖(回应 Gemini 1.0 演示视频争议);Oriol Vinyals权威媒体非本人原话

  • 无法证实I think what the ChatGPT moment did that changed was, and fair play to them to do that, was they demonstrated, I think somewhat surprisingly to themselves as well, that the public were ready to embrace these systems and actually find value in these systems. Impressive though they are, I guess, when we're working on these systems, mostly you're focusing on the flaws and the things they don't do and hallucinations… But interestingly, it turned out that even with those flaws, many tens of millions of people still find them very useful.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://singjupost.com/how-ai-is-unlocking-the-secrets-of-nature-and-the-universe-demis-hassabis-transcript/(在新窗口打开)2024-04TED2024,与 Chris Anderson 对谈;Demis Hassabis权威媒体非本人原话

  • 无法证实Gemini has been mostly top-down, which has been very good in some sense… But it does lead to some incentive to say, 'Hey, what I'm doing is working great.'… There's less of an incentive to say, 'Hey, what I'm doing is not actually working that well.'

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/jeff-dean-and-noam-shazeer(在新窗口打开)2025-02-12Dwarkesh Podcast(论 Gemini 的自上而下体制);Noam Shazeer权威媒体非本人原话

  • 无法证实There's less of an incentive to say, 'Hey, what I'm doing is not actually working that well.'

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/jeff-dean-and-noam-shazeer(在新窗口打开)2025-02-12Dwarkesh Podcast(与 Jeff Dean 同场);Noam Shazeer权威媒体非本人原话

  • 无法证实I mean, I think I actually wrote a one-page memo saying we were being stupid by fragmenting our resources… we were fragmenting not only our compute across those separate efforts, but also our best people. And so I said, this is just stupid.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.latent.space/p/jeffdean(在新窗口打开)2026-02-12Latent Space 播客;Jeff Dean权威媒体非本人原话

  • 无法证实It was a little bit organizationally complicated, and time zone-wise, there were lots of people in London, lots of people here. Eight hours apart is never a recipe for easy collaboration, but I think we've done a really good job of navigating that.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=8hfpLa5wPGo(在新窗口打开)2026-05Gemini co-leads 圆桌(Google for Developers《Release Notes》)· 追述 2023 年合并;Jeff Dean权威媒体非本人原话

  • 无法证实it's never easy bringing two organizations together, but I think everyone realized that this is the right moment… I think both organizations acted with great urgency on that and enabled it.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=8hfpLa5wPGo(在新窗口打开)2026-05Gemini co-leads 圆桌 · 追述 2023 年合并;Koray Kavukcuoglu权威媒体非本人原话

  • 无法证实I don't think there would be big, major disagreements.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=8hfpLa5wPGo(在新窗口打开)2026-05Gemini co-leads 圆桌(被直接问到团队内部分歧时);Koray Kavukcuoglu权威媒体非本人原话

  • 无法证实It didn't just win the match — importantly it actually came up with new creative strategies and ideas that had never been seen before even though we've played Go for thousands of years.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=YtPaZsasmNA(在新窗口打开)2024-12-08诺贝尔化学奖演讲《Accelerating scientific discovery with AI》(斯德哥尔摩)· **口播版,见下方核实范围说明**[@src-5a0e7338dc]讲稿 PDF;Demis Hassabis权威媒体非本人原话

  • 无法证实Innovate architecture of AlphaFold2: a hybrid system (Jumper et al., Nature 2021) — Built in evolutionary and physical constraints into the architecture without impacting the learning / Full end-to-end system with recycling stage that iteratively refines the structure prediction

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.nobelprize.org/uploads/2024/12/hassabis-lecture.pdf(在新窗口打开)2024-12-08诺贝尔奖演讲讲稿**幻灯片**,总结 AlphaFold 的方法论(电报体要点,非口播逐字稿);Demis Hassabis权威媒体非本人原话

  • 无法证实You can see it as a sort of super coding agent… It doesn't just propose a piece of code or an edit, it actually produces a result that maybe nobody was aware of.

    依据:12-quotes §3-5 降级:AlphaEvolve 官方博客不含该引语;补采访链接前不得作逐字引语使用。

    相关来源:https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/(在新窗口打开)2025-05-14AlphaEvolve 发布;Pushmeet Kohli权威媒体非本人原话

  • 无法证实whenever we do a systematic study, on any given problem an AI tool has a success rate of maybe 1% or 2%. It's just that they can buy scale, and you just pick the winners.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/terence-tao(在新窗口打开)2026-03-20Dwarkesh Podcast,章节「Selection bias in reported AI discoveries」(00:30:31);Terence Tao权威媒体非本人原话

  • 无法证实I'd argue we've already had one of those moments with AlphaFold.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=AFpeWo1GTeg(在新窗口打开)2026-04Sequoia AI Ascent 2026 对谈;Demis Hassabis权威媒体非本人原话

  • 无法证实simulations is the way we can address some of what we maybe think of as social sciences, like economics

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=AFpeWo1GTeg(在新窗口打开)2026-04Sequoia AI Ascent 2026 对谈;Demis Hassabis权威媒体非本人原话

  • 无法证实the dream is to do almost all the exploration, which is 99%... in silico

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=AFpeWo1GTeg(在新窗口打开)2026-04Sequoia AI Ascent 2026 对谈,谈 Isomorphic 的终局目标;Demis Hassabis权威媒体非本人原话

  • 无法证实only comes in lightweight flavors

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.dwarkesh.com/p/demis-hassabis(在新窗口打开)2024-02-28Dwarkesh Podcast(解释 Gemma 为什么可以开源);Demis Hassabis权威媒体非本人原话

  • 无法证实we probably don't want to add training objectives which would encourage things like deceit

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.cognitiverevolution.ai/scaling-thinking-gemini-2-5-tech-lead-jack-rae-on-reasoning-long-context-the-path-to-agi/(在新窗口打开)2025-04Cognitive Revolution 播客;Jack Rae权威媒体非本人原话

  • 无法证实Because we are defining the frontier with 2.5 Pro DeepThink, we are taking a little bit of extra time to conduct more frontier safety evaluations and get further input from safety experts. As part of that, we are going to make it available to trusted testers via the Gemini API to get their feedback before making it widely available.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://singjupost.com/transcript-of-google-ceo-sundar-pichai-at-google-i-o-2025-keynote/(在新窗口打开)2025-05-20Google I/O 2025 keynote(解释 Deep Think 为什么延迟发布);Demis Hassabis权威媒体非本人原话

  • 无法证实This model is so good at coding that it triggered our frontier safety early warning trigger for cyber uplift… it's not the critical capability level yet. But it's early warning and we're starting our response plan.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://gist.github.com/simonw/d664f927834e9d687cbd4fd9ffd6b135(在新窗口打开)2025-06-17Gemini 2.5 GA X Space;Anca Dragan权威媒体非本人原话

  • 无法证实something may go wrong and hopefully it's just sort of medium sized

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=ZXA2dmFxXmg(在新窗口打开)2025-12-16Google DeepMind: The Podcast 第 27 集,谈国际治理需要什么才能推动;Demis Hassabis权威媒体非本人原话

  • 无法证实I don't feel like that necessarily anymore

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=ZXA2dmFxXmg(在新窗口打开)2025-12-16Google DeepMind: The Podcast;Anca Dragan权威媒体非本人原话

  • 无法证实It takes up all my mental energy all day, every day.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=ZXA2dmFxXmg(在新窗口打开)2025-12-16Google DeepMind: The Podcast;Anca Dragan权威媒体非本人原话

  • 无法证实① So whatever is designed, and I have some ideas on this and I'll probably be talking about this later this year, it needs to be dynamic, which is a word that doesn't usually go with regulation, right? So it's got to be light, fleet-footed and able to be informed by the latest developments so that it can adapt to where the actual risk is… 〔同一段发言之内,① 与 ② 之间省略 7 句〕 ② And I think some form of really almost smart regulation is required that is dynamic and can adapt with the times very quickly and probably informed by the leading labs because they're seeing what's actually at the coal face.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://singjupost.com/gsb-a-conversation-with-demis-hassabis-ceo-of-google-deepmind-transcript/(在新窗口打开)2026-06Stanford Graduate School of Business 炉边谈话(与院长 Jonathan Levin);Demis Hassabis权威媒体非本人原话

  • 无法证实Our job is to be winning in both cases, right?

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=rq-2i1blAlU(在新窗口打开)2025-11-18NYT《Hard Fork》播客,被问 AI 是否泡沫;Demis Hassabis权威媒体非本人原话

  • 无法证实it's almost an overreaction to the underreaction

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=ZXA2dmFxXmg(在新窗口打开)2025-12-16Google DeepMind: The Podcast 第 27 集(与 Hannah Fry),谈 AI 泡沫;Demis Hassabis权威媒体非本人原话

  • 无法证实We weren't the first company to do search. We weren't the first company to do email. We weren't the first company to build a browser.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=5puu3kN9l7c(在新窗口打开)2026-01Bloomberg 视频访谈;Sundar Pichai权威媒体非本人原话

  • 无法证实We're the only organization that has the full stack — from TPUs and hardware, data centers, the cloud business, the frontier lab, to all these great products.

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://www.youtube.com/watch?v=BbIaYFHxW3Y(在新窗口打开)2026-01-20Bloomberg House @ Davos(Emily Chang 专访);Demis Hassabis权威媒体非本人原话

  • 无法证实我不得不跟很多 Google 内部的团队,比如 legal 团队、operation 团队合作,跟他们说这个时代你是等不起的,你等了 30 天以后,那些非常有使命感的 researcher 都跑了。

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://mp.weixin.qq.com/s/mSn6CAaUiNsZ4-t2VWH73g(在新窗口打开)2025-11-28中文原创长访谈《独家对话 DeepMind 谭捷:机器人、世界模型与 Google》(张小珺 × 谭捷,约 4 万字播客文字版);中文原话谭捷权威媒体非本人原话

  • 无法证实感觉可能比 996 还要更辛苦一点。

    依据:12-quotes 原型入库条目;source_id 已链台账,C09 未逐 URL 打开复核。

    相关来源:https://mp.weixin.qq.com/s/mSn6CAaUiNsZ4-t2VWH73g(在新窗口打开)2025-11-28同上,自述工作强度;中文原话谭捷权威媒体非本人原话