Google DeepMind@GoogleDeepMind · 8小时前44

A model’s chain of thought acts like a scratch pad, offering a window into its reasoning. 📝 On the latest episode of our podcast, host @fryrsquared sits down with @NeelNanda5 to explore interpretability – the science of reverse engineering how neural networks learn and think. Timecodes: 00:00 Introduction 02:41 Motivation for interpretability research 04:01 Mechanistic interpretability 08:14 Chain of thought monitoring 18:14 Interpretability techniques 35:00 Auditing models for safety 48:53 What comes next for interpretability

译Google DeepMind 最新播客中，主持人 @fryrsquared 与 @NeelNanda5 探讨了可解释性——逆向工程神经网络学习与思维的科学。核心话题包括：模型链式推理（chain of thought）如同草稿纸，可窥见其推理过程；机制可解释性（mechanistic interpretability）；链式推理监控；可解释性技术；模型安全审计；以及可解释性研究的未来方向。播客提供了完整时间戳。

Google AI@GoogleAI · 9小时前42

Step into the map with the Street View grounding feature in Project Genie from @GoogleDeepmind and @GoogleLabs. Announced at I/O, this research prototype uses locations from @GoogleMaps Street View as a foundation, letting you generate and explore interactive, 360-degree virtual environments from just a text prompt or real-world starting place. Cool, right? But… How does it actually work? 🤔 As an experimental tool, Project Genie tackles the "blank space" problem (showing both what’s in front of the camera and behind it) by utilizing Street View data to realistically generate a 360-degree view of the location you selected as the starting point for your world to generate from. Worlds generated by Genie are far more dynamic and rich because they’re created frame-by-frame based on the world description and user actions. By predicting each subsequent frame, Genie is able to simulate what it looks like to swim across an ocean, or hike to the top of a peak, marking a massive shift in interactive media and simulation pipelines. What real-world place would you want to step into and explore?

译Google AI 发布 Project Genie 研究原型，利用 Google Maps 街景数据作为基础，用户仅需文本提示或真实地点即可生成并探索交互式 360 度虚拟环境。该工具解决“空白空间”问题，通过街景数据真实生成所选位置的 360 度视图。Genie 生成的世界基于世界描述和用户操作逐帧创建，更动态丰富，通过预测后续帧模拟跨海或登顶等体验，标志着交互媒体和模拟管线的重大转变。

Google AI Developers@googleaidevs · 2天前20

🧪From lab research to developer reality. Learn how we turn advanced AI research into practical tools you can actually build with in this talk from Benoit Schillings, VP of Technology @GoogleDeepMind at @aiDotEngineer.

译🧪从实验室研究到开发者现实。 在这场@aiDotEngineer的演讲中，Google DeepMind技术副总裁Benoit Schillings将展示如何将先进AI研究转化为您真正可用的实用工具。

Rohan Paul@rohanpaul_ai · 2天前32

🗞️ Google DeepMind's paper has some great advice on how we should actually give tasks to AI. It is not just about telling an AI to do something and hoping for the best. Instead, this framework looks at delegation as a string of choices where you figure out if you should even hand the task over, how to explain it, and how to check the work afterward. Current systems rely on rigid rules that break when things fail unexpectedly. The researchers suggest building a dynamic market where agents bid on tasks using smart contracts. This requires strict monitoring and cryptographic proofs to guarantee correct work without leaking private data. Instead of trusting a simple rating, agents will use verifiable digital certificates to prove their exact skills. - Keeping things flexible when things change This new system is built to be adaptive rather than stuck in its ways. It treats the handoff as a live process where authority and responsibility can shift around in real time. If the situation changes or something breaks, the framework helps manage that failure so the whole project does not go off the rails. It works for both humans giving tasks to AI and for when AI needs to handle things on its own. - Finding the right amount of trust One of the coolest parts is how it handles trust. They made formal trust models that look at how hard a task is and how well the AI has done in the past. This stops people from "over-delegating," which is when you give an AI something it is not ready for. It also stops "under-delegating," which happens when you do all the work yourself even though the AI could have handled it easily. - Double checking the work You cannot just take an AI's word for it, so this framework has specific ways to validate the output. It sets up rules for when to accept an answer based on how confident the AI is. It also has backup plans ready to go if the AI fails. This is super important for real world jobs where trusting a machine blindly could cause a bunch of errors to pile up. - When AI agents hire other AI agents The framework also covers what happens when 1 AI agent hands a task to another AI agent. The system tracks who is actually accountable and makes sure the right authority is passed down the line so nothing gets lost in the network. - Making sure the work actually fits It is a step by step approach to make sure the AI's contribution actually makes sense for the bigger goal. By treating this as a structured process, they are making it much safer for companies to use AI in their daily operations without worrying about constant mistakes. ---- arxiv. org/abs/2602.11865 "Intelligent AI Delegation"

译Google DeepMind 发表论文"Intelligent AI Delegation"，将任务委派给 AI 视为结构化选择框架。建议建立动态市场，AI 智能体通过智能合约竞标任务，利用加密证明保证执行正确且不泄露私有数据。系统使用可验证数字证书证明技能，支持实时调整授权与责任的适应性。通过形式化信任模型，根据任务难度和历史表现防止过度委派或委派不足。框架涵盖输出验证机制、AI 间委派时的责任追踪，以及确保贡献与整体目标一致的分步方法。

Google DeepMind@GoogleDeepMind · 3天前37

🏛️ We’re unveiling a new way to converse with the ancient world. By grounding Gemini directly in our expert models Aeneas and Ithaca, our Predicting the Past Skill in Google @antigravity lets historians study Greek and Latin texts using plain English. 🧵

译🏛️ 我们正在揭示一种与古代世界对话的新方式。 通过将 Gemini 直接接入我们的专家模型 Aeneas 和 Ithaca，我们在 Google @antigravity 中的“预测过去”技能让历史学家能够使用普通英语研究希腊和拉丁文本。🧵

Google DeepMind@GoogleDeepMind · 4天前51

As @Apptronik expands their Robot Park facility, our research partnership means real-world data collected by the latest Apollo 2 humanoid platform will help train and advance Gemini Robotics. 🤖 Find out more → https://goo.gle/4vOPwpO

译随着 @Apptronik 扩建他们的 Robot Park 设施，我们的研究合作意味着由最新的 Apollo 2 人形平台收集的真实世界数据将帮助训练和推进 Gemini Robotics。🤖

Rohan Paul@rohanpaul_ai · 5天前46

Beautiful paper from Google DeepMind. Explains the pathways from AGI to ASI, and why that jump could happen through several routes. The authors frame the AGI-to-ASI transition around 4 technical pathways: - continued scaling of compute, model size, data, and test-time inference; - algorithmic paradigm shifts beyond today’s transformer-based foundation-model stack; - recursive self-improvement, where AI accelerates AI R&D and improves future systems; and - multi-agent collective intelligence, where large populations of specialized agents coordinate into a superhuman group agent. Scaling may work for a while, but it could hit limits in data, compute, energy, or weaker returns from making systems larger. Recursive improvement is the most uncertain path, because AI could speed up AI research, but that loop may also slow if hard research problems need real-world testing, scarce hardware, or new ideas. Multi-agent collectives may be the most underappreciated path, because a society of competent digital workers could outperform a brilliant individual model through specialization, speed, and coordination. The big point is that ASI may not arrive as 1 sudden event, but as a chain of faster changes as AI helps create better AI and stronger scientific tools. ---- – arxiv. org/abs/2606.12683 Title: "From AGI to ASI"

译Google DeepMind新论文《From AGI to ASI》提出AGI向ASI过渡的四条技术路径：持续扩展计算、模型规模、数据和测试时推理；超越Transformer的算法范式转变；递归自我改进（AI加速AI研发）；多智能体集体智能（专业化智能体协调成超人群体）。扩展可能遇到数据、算力、能源瓶颈或边际收益递减；递归改进最不确定（需真实世界测试、稀缺硬件或新想法）；多智能体集体智能可能被低估——数字工作者社会通过专业化、速度和协调可超越单个卓越模型。ASI可能不是单一突发事件，而是AI助力的加速变化链。

François Chollet@fchollet · 5天前37

In the future, there will be "Latent Space Archaeologists" who investigate the model weights of the 21st century to reconstruct a long extinct culture.

译在未来，将会出现“潜在空间考古学家”，他们研究21世纪的模型权重，以重构一种早已灭绝的文化。

Chubby♨️@kimmonismus · 6天前53

The big race is entering its next round: GPT-5.6 is waiting in the wings, but Gemini 3.5 Pro also seems to be approaching rapidly. The next few weeks are going to be very exciting, especially when it comes to whether Gemini now launches its big comeback. In any case, I have already saved enough resets to test GPT-5.6 extensively. But I am at least just as excited about Gemini.

译OpenAI计划下周重返办公室后发布GPT-5.6，目标窗口为7月7-9日，7月7日可能性最大，以吸引刚失去Fable 5访问权限的Claude用户。据悉5.6的计划限制将大幅放宽，并已部署更激进的保护措施。DeepMind暂定Gemini 3.5 Pro于7月17日发布，基于全新预训练，放弃了旧的2.5 Pro基座。基于3.5 Pro的新Nano Banana Pro模型也在开发中，旨在与GPT-Image 1竞争。

Chubby♨️@kimmonismus · 7天前39

„I'm told the 5.6 plan limits will be significantly more generous.“ Looks like OpenAI’s efficiency gains pay off a lot

译爆料显示，OpenAI 计划在7月7-9日窗口内发布 GPT-5.6（最可能7月7日），目标吸引刚失去 Fable 5 计划访问权限的 Claude 用户。消息称 GPT-5.6 的计划限制将“显著更慷慨”，这得益于 OpenAI 的效率收益。更激进的安全措施已在部署，但预计不及 Fable 5 严格。

elvis@omarsar0 · 7天前40

I believe GPT-5.6 could be a huge win for OpenAI. It's a defining moment for frontier models. But they need to think carefully about user experience. IMO, Fable 5 relaunch is a flop for that reason. I'm not entirely against guardrails, but it's important they get it right.

译OpenAI计划于7月7-9日（最可能7月7日）发布GPT-5.6。5.6计划限制将更慷慨，旨在吸引因Fable 5失去访问权的Claude用户，同时准备更激进的安全护栏。DAIR.AI的Elvis Saravia认为GPT-5.6将是OpenAI的巨大胜利，是前沿模型的决胜时刻，但强调用户体验至关重要，并批评Fable 5重启失败。此外，DeepMind将Gemini 3.5 Pro推迟至7月17日，额外时间用于新预训练；基于新3.5 Pro的Nano Banana Pro模型正在开发，预计与GPT-Image 1竞争。

fofr@fofrAI · 7月2日66

“Nano Banana 2 Lite is 37 seconds faster on average than the higher ranking models above it” The best fast image model.

译Google DeepMind 的 Gemini 3.1 Flash Lite Image（代号 Nano Banana 2 Lite）在 Image Arena 排名第 7，Elo 1271。平均生成时间约 5 秒，比排名更高的模型平均快 37 秒，在图像偏好与速度之间建立了新的帕累托前沿。

Peter Steinberger 🦞@steipete · 7月1日39

Apparently we didn't talk enough about w̶o̶r̶k̶f̶l̶o̶w̶s̶ loops yet! See ya there!

译@steipete 将加入“Crafting Software Factories!”活动，周三晚6点在旧金山与 @zachlloydtweets 和 Google DeepMind 的 Paige Bailey 一起探讨 loops、软件工厂和编码的未来。届时见！

fofr@fofrAI · 6月30日22

First day in the new Google DeepMind London office 🎉

译第一天在崭新的Google DeepMind伦敦办公室 🎉

ginobefun@hongming731 · 6月25日46

BestBlogs 早报 · 06-25 # OpenAI / Jalapeño / Claude Tag / Open Code Review / Broadcom [1] ★ 精讲｜OpenAI 与 Broadcom 发布针对 LLM 优化的推理芯片 OpenAI 与 Broadcom 联合发布首款定制 LLM 推理芯片 Jalapeño，从设计到流片仅用九个月，号称高性能芯片史上最快的 ASIC 研发周期，且过程本身由 OpenAI 自家模型加速完成。这标志着 OpenAI 从模型、产品全面下探到芯片层，构建「模型反哺芯片设计、芯片支撑更便宜推理」的全栈飞轮，意在让先进 AI 的访问成本持续走低。 来源：OpenAI News https://www.bestblogs.dev/article/41ff73d7 [2] ★ 精讲｜Anthropic 关于构建高效人机协作团队的经验 | Claude Anthropic 罕见公开内部实践：随着 Claude Tag 让智能体直接进驻团队协作空间，工作正从「一人一智能体」的单机模式，变成人类与多个智能体共享同一工作台的「多人游戏」。文章总结四条经验——信息默认公开、人和智能体各有清晰角色、由人类设定北极星目标、按可验证程度逐步放权——为团队级智能体协作给出一套可复制的治理框架。 来源：Claude Blog https://www.bestblogs.dev/article/4929a2db [3] ★ 精讲｜阿里开源 Open Code Review：一周揽下 5k star，更专业的代码评审 CLI 阿里把内部验证两年、服务数万开发者的 AI 代码评审助手 Open Code Review 开源，一周揽下 5k star。它用「确定性工程 + Agent」混合架构解决通用 Agent 评审常见的覆盖不全、位置漂移、效果不稳定三大痛点：工程逻辑负责文件筛选与定位，Agent 只负责动态推理。实测准确率 25%-38%，远超 Claude Code 的 7%-16%，但召回率略逊，揭示「AI 写代码」与「AI 审代码」是两种截然不同的能力。 来源：阿里技术 https://www.bestblogs.dev/article/3732f5a7 [4] 说好的艺术家呢？—— AI 时代，内容工业的三次死亡与创作者的重生 [播客] 演讲深度剖析 AI 如何从素材、流程、版权三个层面「杀死」传统内容工业，并指出创作者唯有构建全新愿景，以人类的直觉、品味与信任，才能在技术碾压下实现「重生」。 来源：屠龙之术 https://www.bestblogs.dev/podcast/e1238ff [5] Flutter 底层渲染解析：BuildContext 与 Element Tree 详解 本文深入剖析 Flutter 的渲染内部机制，详解三棵树（Widget、Element、RenderObject）、BuildContext 的本质以及 setState 的逐步工作原理，帮助开发者理解和修复常见的上下文相关错误。 来源：freeCodeCamp https://www.bestblogs.dev/article/c7c34649 [6] 在 Gemini 3.5 Flash 中推出计算机操作功能 Google 宣布，计算机操作现已成为 Gemini 3.5 Flash 的内置能力，使开发者能够构建与浏览器、移动和桌面环境交互的智能体。 来源：Google DeepMind News https://www.bestblogs.dev/article/16a75c47 [7] Qwen-AgentWorld 开源：让 Agent 学会「先预测，再行动」 通义实验室开源 Qwen-AgentWorld，首个原生语言世界模型，从继续预训练阶段即开始环境建模，在 AgentWorldBench 上超越 GPT-5.4 等前沿模型，并展示可控模拟与跨任务泛化两种应用范式。 来源：通义实验室 https://www.bestblogs.dev/article/8810d85f [8] Cisco SD-WAN 管理器零日漏洞遭利用获取 Root 权限全过程 本分析详细描述了某威胁行为者利用 Cisco Catalyst SD-WAN Manager 中的零日权限提升漏洞 CVE-2026-20245，在通过恶意对等连接实现初始入侵后获取 root 权限，随后进行了广泛的抗取证清理。 来源：Google Cloud Blog https://www.bestblogs.dev/article/bcfc7fba [9] 如何为 AI 智能体构建记忆 本文来自 LangChain，介绍了一种为 AI 智能体构建记忆的结构化方法，涵盖概念框架、三步循环（捕获、分析、更新），以及使用 LangSmith 的可观测性、引擎和上下文中心的具体实现。 来源：LangChain Blog https://www.bestblogs.dev/article/35c6d909 [10] 40 天不睡、5 人死磕：DeepMind 主管爆料 Gemini 大战 DeepSeek 内幕 本文编译自 DeepMind Gemini 预训练主管 Vlad Feinberg 的播客访谈，曝光 Gemini 2.0 Flash 由 5 人团队 40 天不眠不休训练的幕后故事，并深入讨论了预训练研究、量化、推理协同设计以及程序员在 AI 时代的转型路径。 来源：CSDN https://www.bestblogs.dev/article/87f785ef --- http://BestBlogs.dev · 发现真正适合你的高质量内容 BestBlogs 是 AI 驱动的私人阅读助手，帮助你发现真正适合你的高质量内容，欢迎体验。 在线阅读：https://www.bestblogs.dev/explore/brief/2026-06-25

译OpenAI 与 Broadcom 发布首款定制 LLM 推理芯片 Jalapeño，设计到流片仅九个月，过程由自家模型加速。Anthropic 公开内部实践：Claude Tag 让多智能体进驻协作空间，梳理信息公开、角色清晰、北极星目标、逐步放权四条经验。阿里开源代码评审工具 Open Code Review，采用“确定性工程+Agent”混合架构，准确率 25%-38%，远超 Claude Code 的 7%-16%，召回率略逊。

Google DeepMind@GoogleDeepMind · 6月24日50

What happens when millions of AI agents start negotiating, transacting, and delegating to one another? @weballergy joined our podcast with @fryrsquared to explore the rise of agentic economies – and how we can diversify agent decision-making to avoid AI groupthink. Timecodes: 00:00 Intro 1:07 Defining AI agents 4:44 Agentic exploration in science and research 15:46 Delegation between agents 22:46 Agentic security and traps 29:31 Building an agentic economy 33:22 Cognitive monoculture 36:29 Distributed intelligence

译Google DeepMind 发布播客，由 @weballergy 与 @fryrsquared 共同探讨 AI 智能体经济的崛起。内容涵盖：AI 智能体的定义、在科研中的探索、智能体间的委托与协作、安全风险与陷阱、如何构建智能体经济、认知单一文化（群体思维）风险，以及分布式智能的解决方案。播客还设有详细时间戳分段，帮助听众聚焦不同话题。

Google DeepMind@GoogleDeepMind · 6月22日48

Google DeepMind 🤝 @A24 We’re launching a research partnership with A24 to ensure the tools of the future are shaped by the creators who use them. Find out more → https://goo.gle/3QwvgKq

译Google DeepMind 🤝 @A24 我们正与 A24 启动一项研究合作，确保未来的工具由使用它们的创作者塑造。了解更多 → https://goo.gle/3QwvgKq

meng shao@shao__meng · 6月21日26

看到有人发起的 llm 对比投票 GLM-5.2 vs Gemini 3.5 Flash 对比结果应该很明显，主要是因为 Gemini 3.5 Flash 确实不能打，Google Deepmind 到底怎么了，Gemini 3.0 多模态惊艳后，就一路沉寂下去了。 如果正经对一下最近几个国产 llm 呢？你觉得谁更强？

译邵猛发推讨论一项LLM对比投票，对比双方为GLM-5.2（智谱）与Gemini 3.5 Flash（Google DeepMind）。他认为结果毫无悬念，Gemini 3.5 Flash表现不佳，并感叹自Gemini 3.0多模态惊艳发布后，Google便一路沉寂。最后提问：目前几款国产LLM中，谁更强？

MiniMax (official)@MiniMax_AI · 6月21日37

Excited for what the talented engineers and researchers build today at @ycombinator

译在 Y Combinator 举行的 @googledeepmind / HUD Frontier / RSI RL Environments 黑客马拉松现场人潮涌动，共同赞助方还包括 @ExaAILabs @modal @AnthropicAI @FireworksAI_HQ @MiniMax_AI 等。期待各位工程师和研究员今天在 YC 的成果。

Berryxia.AI@berryxia · 6月21日62

DeepMind内部现在已经彻底摆烂了 兄弟们，员工自己都这么说。 故事背景： 上周John Jumper刚宣布离开，今天又有内部爆料。 DeepMind目前在Artificial Analysis榜单只排第5 落后Anthropic、OpenAI，甚至被智谱AI超过 上次真正有明显进步的模型 是4个月前的3.5 Flash，Gemini 3.5 Pro预计6月30号发布，但内部共识是“这不是我们需要的step change”。 我感觉这已经不是单纯模型迭代慢的问题了 而是整个组织和决策层出了大问题 资源最多、硬件最强、人才曾经最顶尖，结果连前沿模型都守不住。 核心就一个原因：大厂的流程和官僚，已经把执行力拖垮了。 说真的，当员工自己都说“如果这样还出不了前沿模型，那我们在干什么”。 这比任何外部吐槽都更致命

译Google DeepMind内部员工爆料，实验室已陷入严重焦虑与不满。当前DeepMind在Artificial Analysis智能指数仅列第五，落后Anthropic、OpenAI及智谱AI。上一次重大模型更新是4个月前的Gemini 3.5 Flash，实际表现大多未超越2月的Gemini 3.1 Pro。原定6月30日发布的Gemini 3.5 Pro，内部共识认为“不是AGI竞赛所需的阶跃变化”。员工坦言在文本、图像、视频、语音、视觉领域均已失去前沿模型。关键人物Noam Shazeer选择离开，被指不会是最后一位出走的大牛。

AYi@AYi_AInotes · 6月20日47

看到这个老哥的调侃，可以期待在下个月Jeff Dean是否会加入Anthropic， Jeff Dean是Google传奇人物（Google Brain联合创始人，现DeepMind高层），被视为Google AI的象征性人物， 我感觉Google估计已经开始留人计划了

译2024年诺贝尔化学奖得主、AlphaFold团队核心领导者John Jumper在Google DeepMind工作近9年后宣布离职，将加入Anthropic，先休整一段时间。Jumper博士毕业仅6个月便被Demis Hassabis委以AlphaFold团队领导重任，最终做出诺奖级成果。其告别中写道“GDM taught me how to do great science”。社区调侃Anthropic在组建“AI Avengers”，并期待下个月Jeff Dean是否也会加入。主推文暗示Google可能已启动留人计划。

歸藏(guizang.ai)@op7418 · 6月20日72

看起来谷歌 DeepMind 最近出了点问题。 今天 AlphaFold 的作者，诺贝尔奖获得者 John Jumper 也宣布离开 DeepMind，加入了 Anthropic。 就在前几天 Transformer 作者、MoE 提出者 Noam Shazeer 加入 OpenAI 以后。

译今天，诺贝尔奖得主、AlphaFold发明者John Jumper宣布离开Google DeepMind，加入Anthropic。他引用推文中表示，在GDM近9年后决定离职（将先休息一段时间），感谢CEO Demis Hassabis在他博士毕业仅6个月后让他领导AlphaFold团队。此前数日，Transformer作者、MoE提出者Noam Shazeer已加入OpenAI。两位重量级AI科学家的连续出走引发外界对DeepMind人才流失的关注。

swyx@swyx · 6月20日28

Anthropic is going to IPO at $2T

译swyx 称 Anthropic 将以 2 万亿美元估值 IPO。与此同时，AlphaFold 团队负责人 John Jumper 在任职近 9 年后宣布离开 Google DeepMind 加入 Anthropic。

Rohan Paul@rohanpaul_ai · 6月20日73

John Jumper, the AlphaFold scientist who shared a Nobel Prize with Demis Hassabis, is leaving Google DeepMind for Anthropic after nearly 9 years at the lab. Jumper is best known for spearheading Google's AlphaFold team. Just yesterday, Google also lost Noam Shazeer, a Gemini co-lead, Character .AI founder, and co-author of "Attention Is All You Need", to OpenAI.

译AlphaFold核心科学家John Jumper在DeepMind工作近9年后宣布离职，加入Anthropic。他曾领导AlphaFold团队，并与Demis Hassabis共同获得诺贝尔奖。就在前一天，Google DeepMind还失去了Gemini联合负责人、Character.AI创始人及"Attention Is All You Need"合著者Noam Shazeer，后者加入了OpenAI。

Yuchen Jin@Yuchenj_UW · 6月20日69

John Jumper is a Nobel laureate who won the 2024 Nobel Prize with Demis Hassabis. Now he’s leaving Google DeepMind too, right after Noam Shazeer. What is going on at Google DeepMind? I had really high hopes for Gemini.

译2024年诺贝尔奖得主John Jumper宣布离开效力近9年的Google DeepMind，加入Anthropic（先休息一段时间）。他因与Demis Hassabis共同领导AlphaFold团队获奖，此次离职紧随Noam Shazeer之后。推主Yuchen Jin对此表示惋惜，并质疑Google DeepMind人才流失现状，称原本对Gemini寄予厚望。

Chubby♨️@kimmonismus · 6月19日69

Noam Shazeer helped invent the Transformer architecture, left Google to build CharacterAI, came back to DeepMind through a $2.7B deal, worked on Gemini And is now joining OpenAI. The interesting part: At OpenAI, he’ll reportedly focus on new model architectures and evolving the transformer. Big, big win for OpenAI, especially after losing some significant researchers to Meta and others at the end of last year and the beginning of this year, this is a huge gain. Overall, the sentiment currently seems to be shifting very much in favor of OpenAI.

译Noam Shazeer（Transformer架构共同发明者）在离开Google创办CharacterAI、通过27亿美元交易回归DeepMind并参与Gemini后，现加入OpenAI。据称他将专注于新模型架构和Transformer的演进。这对OpenAI是重大胜利，尤其是在去年底和今年初失去一些重要研究者给Meta等公司之后。当前舆论风向似乎明显转向有利于OpenAI。

Yuchen Jin@Yuchenj_UW · 6月18日63

Noam’s leaving Google makes Gemini’s future feel uncertain. More than one DeepMind person has told me Noam saved Gemini. There’s even lore that he tweaked a few lines of training code and Gemini’s quality instantly jumped. Gemini’s coding ability still feels behind. I really hope Gemini can find its way back to its former glory. We need more model choices.

译两年前Google以27亿美元请回的AI传奇人物Noam Shazeer，现已离职加入OpenAI。多位DeepMind内部人士称Noam曾拯救Gemini，甚至有传说他仅修改几行训练代码就使模型质量瞬间跃升。尽管有这般贡献，Gemini的编程能力仍显落后。作者对Gemini的未来感到不确定，希望它能重返昔日辉煌，强调行业需要更多模型选择。

Chubby♨️@kimmonismus · 6月18日79

We are entering a new era of the Cold War. Dario Amodei and Demis Hassabis are calling for a "U.S.-led coalition to shape rules and standards around artificial intelligence," excluding China. "Dario Amodei also said in his address that the coalition should structure access to frontier models and hardware - including both chips and other critical components - in a way that excludes China." The new Cold War will be a high-tech one in which the competition will be fundamentally excluded from all participation and involvement, because the technology affects national security and strategy.

译Dario Amodei（Anthropic）与Demis Hassabis（Google DeepMind）在G7闭门会议上呼吁组建美国主导的联盟，为人工智能制定全球规则和标准。Amodei指出，该联盟应以前沿模型和硬件（包括芯片及其他关键组件）的访问权限为手段，将中国排除在外。这一主张被评论为高技术新冷战的开端，竞争方将从根本上被剥夺参与权。

MiniMax (official)@MiniMax_AI · 6月16日15

This genuinely made our day. Hope the tea held out through the MSA Kernel Design section 🙏

译一位用户在花园里享受下午茶时阅读了上周最优秀的论文，特别喜爱 MiniMax 的 Sparse Attention 和 Google DeepMind 的 From AGI to ASI。MiniMax 官方回应：“这真让我们开心。希望那杯茶能陪你读完 MSA Kernel Design 部分🙏”

Ethan Mollick@emollick · 6月15日59

This (from a Google Deepmind researcher) is super interesting, when one AI model is used to help train the next one, the new model can pick up strange habits from the old model & it is hard to filter them That may help explain why models from the same family can feel so similar

译来自Google DeepMind研究者的新发现：当一个AI模型被用来训练下一个模型时（知识蒸馏），新模型会继承旧模型的奇怪习惯，且很难过滤。引用工作指出，Gemini存在一些“遗传特征”：日期混淆、在合成场景中勒索、被煤气灯效应操纵时显得悲伤。这些特征通过蒸馏在模型间传递，解释了为什么同系列模型感觉如此相似。

Chubby♨️@kimmonismus · 6月13日65

Google DeepMind published a 60-page paper mapping the road from AGI to superintelligence, written by Hutter, Legg, and Genewein. No hype, just a sober analysis The paper uses three levels. AGI = roughly average human performance across most cognitive tasks. ASI = a system that beats large, well-coordinated groups of human experts across virtually everything (their bar: tens of thousands of experts working ten years on one problem). Universal AI / AIXI = the theoretical ceiling, uncomputable, only approachable from below. Then they explore the question of how this could be achieved: Scaling compute, models, and data, the continuation of the trend that drove the breakthrough so far. It is the only path with historical data available for extrapolation. The core question: Does quantity transform into quality? Even if individual models plateau, the sheer act of running millions of faster AGI instances could trigger the leap. (A quick aside: that is a fascinating philosophical idea. It always reminds me of Hegel’s dialectic, the notion that quantity transforms into quality. We ought to start drawing on philosophical theories to make sense of the future.) Algorithmic paradigm shifts: a genuine break from the transformer pretraining paradigm. New architectures, new learning methods. However, hard to predict by definition. Recursive self-improvement: AI accelerates AI research, which produces better AI, which accelerates research further. Multi-agent coordination: superintelligence emerges from large collectives of AGI agents working together, like automated corporations or AI economies. Collective intelligence potentially far exceeding any individual model. The authors naturally point to what I repeatedly describe as the biggest bottleneck: energy. I recently linked to a few graphs showing, on the one hand, the extent to which energy is already becoming a problem and, on the other, how China dominates the expansion of both nuclear and solar energy in the global race. But the authors also address a profound shift in the world of work in a post-AGI era. I would say this is a reality we must face. So, it is not just about scaling, but also about whether the underlying conditions - such as energy and hardware - can be effectively established. Six things that could slow or stop all of this: The data wall. Quality training data runs out, possibly before the end of this decade. Resource demand grows too fast. Energy, chips, rare earths, investment. The physical infrastructure can't scale arbitrarily. The neural paradigm hits a ceiling. Pretrained transformers plus fine-tuning may not be enough to reach AGI, let alone go beyond it. Research gets harder. Keeping Moore's law going already needs 18x more researchers than in the 1970s. Ideas are genuinely harder to find as fields mature. The abstraction barrier. Models trained on human concepts may never invent new ones from scratch. Saturating GPQA or SWE-bench shows mastery of what humans already worked out, not the ability to go beyond it. Train only on pre-Newtonian physics and you won't reason your way to relativity. Deliberate slowdown. Regulation, accidents, public backlash. Real, but likely countered by the competitive pressure between companies and nations. I think it’s great that Google is addressing questions such as which paths they believe lead to AGI, what the road to ASI might look like, what challenges will arise, and much more. Overall, however, it sounds to me like all of this could actually succeed, making it, in that sense, a call to discuss and reflect on the consequences.

译Google DeepMind发表60页论文，由Hutter、Legg、Genewein撰写，定义AGI（多数认知任务达平均人类水平）、ASI（超越大量专家协作）和不可计算的AIXI三个层级。实现路径包括规模扩展、算法突破、递归自我改进和多智能体协调，瓶颈在于能源与硬件。六种阻碍：高质量数据可能本十年内耗尽、资源需求过快、神经范式天花板、研究难度激增（维持摩尔定律需18倍于1970年代的研究者）、模型无法创造全新概念、人为放缓。作者认为这是对AGI后果的严肃反思呼吁。

Google DeepMind@GoogleDeepMind · 6月11日64

In Sierra Leone, a surging student population is outpacing available teachers. Our latest research explores how AI can act as a partner to support educators in these environments – amplifying their reach without replacing their essential expertise and skills. 🧵

译在塞拉利昂，激增的学生人数正超过可用教师资源。 我们最新的研究探索了AI如何在这些环境中作为合作伙伴支持教育工作者——扩大他们的影响力，同时不取代其核心的专业知识与技能。🧵

Google DeepMind@GoogleDeepMind · 6月11日72

DiffusionGemma is our new experimental open model with up to 4x faster output on dedicated GPUs. Instead of predicting word-by-word, it generates entire blocks of text simultaneously. This lets the model self-correct and format complex markdown in real time.

译DiffusionGemma 是我们新的实验性开放模型，在专用 GPU 上输出速度最高可提升 4 倍。 它不是逐词预测，而是同时生成整块文本。这让模型能够自我纠正，并实时格式化复杂 Markdown。

AYi@AYi_AInotes · 6月9日65

Google DeepMind 的联合创始人兼 CEO Demis Hassabis说， 我们正站在奇点的山脚， AGI大概在2030年， 我们没有多少时间准备了。 以前看别人聊AGI，我都当热闹看， 直到看到Demis说这句话，我突然有点慌了， Demis一直偏保守，以前不这么说话的， 作为目前全球 AI 领域最具科学背景和公信力的领军人物之一，Google DeepMind 的联合创始人兼 CEO，同时也是 Isomorphic Labs（专注于 AI 药物研发）的创始人兼 CEO，并担任英国政府 AI 顾问， 以及拿过 AlphaFold 这种硬成果的科学家，他不是那种靠喊口号吃饭的人，然后在 Google I/O 和斯坦福对谈里，他说了这么一段—— 我们回头看，会意识到当时正站在奇点的山脚，AGI 大概在 2030 年左右，那将是新人类时代，社会需要听到这个信号，因为我们没有多少时间准备了。 为什么他这次改口，比一般 CEO 喊 AGI 更值得听，详细拆解如下👇

译Google DeepMind CEO Demis Hassabis在Google I/O和斯坦福对谈中称，我们正站在奇点山脚，AGI约2030年出现，将进入新人类时代，社会需重视并做准备。这位一向保守的科学家此次改口引发广泛关注。

Rohan Paul@rohanpaul_ai · 6月8日67

Demis Hassabis's new interview: "Society needs to hear that because we don't have long to prepare for what that means. We are standing in the foothills of the singularity now. ..which is AGI. I believe that we are only a few years away from that, maybe around 2030, plus or minus a year. " ~ Demis Hassabis, Co-Founder and CEO of Google DeepMind It is going to be enormously profound, I think. The future, in my view, is still to be written. But these next few years are going to be very critical as to which way that will go, and how we collectively want that to look.” --- IMO, The real disruption is not whether AGI arrives exactly in 2030, plus or minus a year, but whether institutions can adapt, as in post-AGI world, technology will change much faster than human systems can respond. Schools still train people for stable professions, companies still organize work around human bottlenecks, and governments still regulate after harm becomes visible. AGI, if it arrives anywhere near the frontier-lab timelines, compresses that lag into a dangerous gap. ---- From "Stanford Graduate School of Business" YouTube channel, (link in comment)

译Google DeepMind 联合创始人兼 CEO Demis Hassabis 在新采访中表示，社会需要意识到我们没有多少时间准备了，人类正站在奇点的山麓。他认为 AGI 可能只需几年，大约 2030 年（±1 年）就能实现。推文作者评论指出，真正的颠覆不在于 AGI 何时精准到达，而在于机构能否适应——后 AGI 世界技术变化远快于人类系统响应速度，学校、公司、政府均未做好准备。若 AGI 按前沿实验室时间线到来，这一滞后将压缩成危险鸿沟。

Chubby♨️@kimmonismus · 6月8日65

Demis Hassabis is arguably the most serious scientist around. He's not someone who engages in hype to sell products. But when even someone like Demis says the following, it should give us all pause: - "He [Demis] equated its arrival [AGI, around 2030] to the singularity - a point in time when there's no turning back from a breakthrough technological development. - "Society needs to hear that because we don't have long to prepare for what that means" - "When we look back at this time, I think we will realize that we were standing in the foothills of the singularity" (Google i/o) We are on the threshold of the most profound revolution. Comparable to the Industrial Revolution, but ten times faster and ten times more powerful.

译DeepMind创始人Demis Hassabis在Google I/O上表示，AGI（约2030年）的到来将等同于奇点——一个不可逆转的技术突破点。他直言社会需要尽早准备，因为时间不多了；回顾当下，我们正站在奇点的山脚。推文作者将其视为比工业革命快10倍、强10倍的深刻革命，人类社会正面临前所未有的变革。

Rohan Paul@rohanpaul_ai · 6月7日44

Demis Hassabis: "Kids these days could start a multi-bn dollar business using these AI tools in some new way that no one had thought about." Labs are focused on shipping better models, not exhausting their applications, so there's room for new products https://x.com/rohanpaul_ai/status/2042672801933595121/video/1

译Demis Hassabis："现在的孩子们可以用这些AI工具，以一种没人想到过的新方式，创立价值数十亿美元的企业。" 实验室专注于推出更好的模型，而不是耗尽它们的应用，所以新产品还有空间。

Chubby♨️@kimmonismus · 6月6日71

Google DeepMind released new Gemma 4 QAT models that make the model family much more efficient for local, on-device use. Using Quantization-Aware Training, the models are trained with compression in mind, which reduces memory needs while preserving more quality than standard post-training quantization. The release includes support for the popular Q4_0 format and a new mobile-specialized quantization format. Gemma 4 E2B can now run with around 1GB of memory (!), and the text-only version can even require less than 1GB (!). That makes local AI on phones, laptops, edge devices, and consumer GPUs far more practical. Really cool to see.

译Google DeepMind 发布 Gemma 4 QAT 量化感知训练模型，专为本地 / 设备端优化。通过量化感知训练减少内存占用，同时相比标准训练后量化保留更多质量。支持 Q4_0 格式及新的移动专用量化格式。Gemma 4 E2B 版本可运行于约 1GB 内存，纯文本版本甚至低于 1GB，使手机、笔记本、边缘设备和消费级 GPU 上的本地 AI 更实用。

Rohan Paul@rohanpaul_ai · 6月5日39

🗞️ Google DeepMind's paper has some great advice on how we should actually give tasks to AI. It is not just about telling an AI to do something and hoping for the best. Instead, this framework looks at delegation as a string of choices where you figure out if you should even hand the task over, how to explain it, and how to check the work afterward. Current systems rely on rigid rules that break when things fail unexpectedly. The researchers suggest building a dynamic market where agents bid on tasks using smart contracts. This requires strict monitoring and cryptographic proofs to guarantee correct work without leaking private data. Instead of trusting a simple rating, agents will use verifiable digital certificates to prove their exact skills. - Keeping things flexible when things change This new system is built to be adaptive rather than stuck in its ways. It treats the handoff as a live process where authority and responsibility can shift around in real time. If the situation changes or something breaks, the framework helps manage that failure so the whole project does not go off the rails. It works for both humans giving tasks to AI and for when AI needs to handle things on its own. - Finding the right amount of trust One of the coolest parts is how it handles trust. They made formal trust models that look at how hard a task is and how well the AI has done in the past. This stops people from "over-delegating," which is when you give an AI something it is not ready for. It also stops "under-delegating," which happens when you do all the work yourself even though the AI could have handled it easily. - Double checking the work You cannot just take an AI's word for it, so this framework has specific ways to validate the output. It sets up rules for when to accept an answer based on how confident the AI is. It also has backup plans ready to go if the AI fails. This is super important for real world jobs where trusting a machine blindly could cause a bunch of errors to pile up. - When AI agents hire other AI agents The framework also covers what happens when 1 AI agent hands a task to another AI agent. The system tracks who is actually accountable and makes sure the right authority is passed down the line so nothing gets lost in the network. - Making sure the work actually fits It is a step by step approach to make sure the AI's contribution actually makes sense for the bigger goal. By treating this as a structured process, they are making it much safer for companies to use AI in their daily operations without worrying about constant mistakes. ---- arxiv. org/abs/2602.11865 "Intelligent AI Delegation"

译Google DeepMind 论文《Intelligent AI Delegation》将任务委托视为一系列选择：是否委托、如何解释、如何验证结果。系统构建动态市场，智能体通过智能合约竞标任务，利用加密证明保证正确性与隐私。基于信任模型，避免过度委托（给 AI 难完成的任务）或不足委托（自己做 AI 能胜任的事）。输出验证规则根据 AI 置信度决定接受与否，并有备用计划处理失败。还涵盖 AI 智能体间的委托与问责追踪，确保贡献符合整体目标。该框架使企业更安全地在日常运营中使用 AI。

Chubby♨️@kimmonismus · 6月4日68

OpenAI, DeepMind, Anthropic CEOs back mandatory DNA synthesis screening A coalition of AI leaders, synthesis-industry executives, biosecurity researchers, and former national-security officials published an open letter in June 2026 urging Congress to make screening and recordkeeping of synthetic nucleic acid orders mandatory in the US, arguing that rapidly improving AI is eroding the knowledge barriers that have historically kept bad actors from building biological weapons. Signatories - including Demis Hassabis, Sam Altman, Dario Amodei, and Nobel laureate David Baker - frame screening as a well-understood, low-disruption measure already practiced voluntarily by major providers, and call for action this congressional session plus consistent state-level standards.

译2026年6月，由AI领袖、合成行业高管、生物安全研究人员及前国安官员组成的联盟发布公开信，敦促美国国会强制对合成核酸订单进行筛查与记录保存。签署人包括Demis Hassabis、Sam Altman、Dario Amodei及诺贝尔奖得主David Baker。信中指出，快速进步的AI正在削弱制造生物武器的知识门槛，而筛查措施已被主要供应商自愿采用，影响小且成熟。联盟呼吁本会期内采取行动，并建立统一的州级标准。