Google DeepMind

179: 蒸馏风暴：一场无人公开谈论的技术竞赛 adds Google DeepMind as one of the closed frontier labs whose user agreements broadly restrict using model outputs to improve competing models. The episode does not center Google DeepMind as an accuser, but it places Gemini-family access inside the same AI Model Distillation Governance and Frontier Model Access Restrictions frame as Anthropic and OpenAI.

咖啡豆｜「和牛自由」成自助餐厅卖点，贵价光环从何而来？ adds a source-reported organizational update. The episode says Demis Hassabis will move away from daily operations into chair and chief-scientist roles, with management moving toward the chief technology officer, while Jeff Dean and other senior researchers are associated with a new self-improving-AI startup that Google invests in. The wiki treats this as a source-scoped operating-speed and governance update, not as an independently verified corporate record.

E245｜藏在大模型背后的新闻人：GPT们的回复是这样写出来的 adds Google DeepMind as the later workplace for 东尼 / Tony, whom the source describes as a senior model designer after earlier media and Meta content-internationalization work. The episode’s contribution is not a research milestone but a role boundary: frontier labs also need people who can define answer quality, conversation depth, examples, and cultural fit through Content Engineering and AI Answer Evaluation.

E226｜聊聊DeepMind创始人哈萨比斯：一个科学家与失控的AI竞赛 adds the pre-Google history behind this organization through DeepMind and Demis Hassabis. The source traces the lab from Reinforcement Learning AGI Path and unusual AGI financing into Google’s acquisition, AlphaGo, AlphaFold, and the 2023 consolidation with Google Brain after ChatGPT and Bard pressure made Gemini catch-up urgent.

Google DeepMind is discussed in EP108 Vibe Coding大地震：Cursor定价争议、Windsurf收购风波，模型厂商亲儿子们又将如何进场？ as the destination for Windsurf’s CEO and core employees. The episode treats the move as part of the same Model Provider Tool Competition pattern as Claude Code and Gemini CLI: model providers want direct access to the coding workflow rather than leaving the category entirely to editor startups.

134. 【数据的综述】和谢晨聊，新时代的石油、历史、版图、数据金字塔、定价与Recipe adds Google DeepMind as one of the major model organizations that 谢晨 expects to matter in Embodied AI, World Models, and physical-AI data demand.

140. 对姚顺宇的4小时访谈：请允许我小疯一下！在Anthropic和Gemini训模型、技术预测、英雄主义已过去 adds Google DeepMind as Yao Shunyu / 姚顺宇’s post-Anthropic home and as the organization behind his current ML Coding and Long-Horizon AI work. In his account, Google DeepMind offers broad research freedom and many technical directions, while Gemini’s progress depends on clearer organization, engineering execution, and product integration rather than a single heroic researcher.

137. 对洪乐潼的4小时访谈：AI for Math、把数学变成Lean、数学天书中的证明、直觉、被创造与被发现的 adds Google DeepMind as a central AI For Math precedent through AlphaGeometry and AlphaProof. Hong Letong / 洪乐潼 treats DeepMind’s 2024 IMO progress as a field milestone, while contrasting Axiom’s system choices with AlphaProof-style search costs.

AI4S 需要狂人与野心家｜对话英灵殿 Odin："如果神存在，我怎能容忍自己不是神？"【公路播客】 adds AlphaFold as an AI-for-biology precedent associated with Google DeepMind in the broader wiki. Haotian Odin / 浩天 treats AlphaFold as evidence that modern AI architectures can transform protein work, while using it as a stepping stone toward AI Protein Design and All-Modal Molecular World Model rather than as the final shape of scientific AI.

170: 【具身季报 26Q2】世界模型大风不停，和不想被贴标签的人 adds Google DeepMind Robotics through ER 1.6. The source frames ER 1.6 as stronger spatial and task understanding rather than a direct robot-control policy, and reads Google’s partnerships with robot bodies such as Boston Dynamics and Apptronik-style platforms as a possible “Robotics Android” route.

How convergence will define the tech sector in 2026 adds a public forecasting example: Amy Webb cites Google DeepMind training a robot to tie a shoe in 2024. The episode uses the example to show why apparently simple human tasks can reveal the hard part of Physical AI: contextual understanding, physical-world data, and adaptive action.

E244｜端到端vs上下分层：机器人路径之争，正在转向？ adds Han Zheng / 韩正’s view that Google DeepMind plus Boston Dynamics may be one of the strongest robotics combinations over the next two to three years. The source emphasizes Google’s possible advantages in simulation, World Models, compute, talent, and organizational commitment when paired with an advanced robot body.

Sam Altman on YC, OpenAI, and the Meaning of Formidable adds Sam Altman’s claim that DeepMind leadership did not believe in the language-model approach at the time OpenAI committed to it. The source uses that contrast to explain why OpenAI converted the transformer architecture into the Language Model Scaling Bet while Google and DeepMind made different technical and organizational choices.

Connections

• 东尼 / Tony, Content Engineering, AI Answer Evaluation, and AI Interaction Internationalization — E245’s model-design and media-judgment extension.
• Windsurf — talent and strategic transaction case.
• Gemini and Gemini CLI — Google-linked model and coding interface discussed in the source.
• OpenAI and Anthropic — frontier AI peers in the model-provider tool competition.
• Model Provider Tool Competition — concept this source uses Google DeepMind to illustrate.
• Embodied AI, World Models, Data Recipe Co-Creation, and Robotics Simulation Evaluation — physical-AI context added by episode 134.
• Yao Shunyu / 姚顺宇, ML Coding, Long-Horizon AI, and AI Organization Design — frontier-training and organization context added by episode 140.
• AlphaGeometry, AlphaProof, AI For Math, and Axiom — AI-for-math comparison context added by episode 137.
• AlphaFold, AI Protein Design, and AI For Science — AI-for-biology precedent added by the Shizilukou Crossing source.
• Embodied AI, Physical AI, Vision Language Action Models, and World Model VLA Fusion — robotics-brain and ecosystem context added by the LateTalk source.
• Amy Webb, AI Convergence, and Physical AI - Marketplace Tech forecast using shoe tying as an embodied-AI difficulty signal.
• Boston Dynamics, Sim2Real, and Robotics Simulation Evaluation — robotics hardware and simulation comparison added by E244.
• OpenAI, Google, Sam Altman, and Language Model Scaling Bet - language-model route contrast added by The Social Radars.
• DeepMind, Demis Hassabis, Reinforcement Learning AGI Path, and Scientific Ideal vs AI Arms Race — original DeepMind history and AI-race tension added by Silicon Valley 101.
• Jeff Dean, Google, and AI Commercialization Pressure - organization-speed and self-improving-AI startup update added by 声动早咖啡.
• Model Distillation / 模型蒸馏, AI Model Distillation Governance, Frontier Model Access Restrictions, and Gemini - LateTalk episode 179’s closed-model ToS and provenance branch.