The Download：谷歌AI大调整与Meta失控模型

<p><em>这是今天的</em><a href="https://forms.technologyreview.com/newsletters/briefing-the-download/?_ga=2.179569122.736533416.1649661040-405833893.1649413289"><em>The Download</em></a><em>，我们的工作日通讯，每天提供科技世界动态。</em><br></p> <h3 class="wp-block-heading"><strong>谷歌的AI帝国正在重塑。以下是变化情况。</strong></h3> <p>在<a href="https://www.businessinsider.com/google-ai-talent-wars-anthropic-jumper-shazeer-karpathy-openai-2026-6">科技人才争夺战中的惨痛损失</a>、<a href="https://www.cnbc.com/2026/07/16/alphabet-stock-gemini-3-5-pro-ai.html">下一代旗舰模型的延迟发布</a>，以及<a href="https://www.axios.com/2026/07/23/googles-deep-mind-ai-model-race">关于员工士气低落的传闻</a>之后，谷歌宣布对其AI业务进行重大调整。以下是事件概览及其重要性。</p> <p><strong>DeepMind首席执行官德米斯·哈萨比斯将不再负责该部门的日常运营</strong><br>他将出任<a href="https://www.axios.com/2026/08/05/google-deepmind-demis-hassabis-ai">该部门董事长及Alphabet首席科学家</a>，这是一个更广泛的角色。DeepMind现将由首席技术官Koray Kavukcuoglu领导，头衔为高级副总裁。</p> <p><strong>DeepMind可能并入谷歌更广泛的业务</strong><br>谷歌预计将<a href="https://www.ft.com/content/61d41764-f2f7-4906-a112-ff3073972c51?syn-25a6b1a6=1">加强对这个12年前收购的AI实验室的控制</a>。</p> <p><strong>杰夫·迪恩离职创办新的AI公司<br></strong>在谷歌工作27年后，该公司前首席科学家正与三位前同事<a href="https://www.nytimes.com/2026/08/05/technology/google-researchers-ai-startup.html">共同创办Discovery Loop</a>。这家初创公司的目标是实现科研过程的完全自动化，谷歌是其早期投资者之一。</p> <p><strong>这些变化发生在谷歌AI面临财务担忧之际</strong><br>最近一个季度，该公司<a href="https://www.cnbc.com/2026/08/05/google-chief-scientist-jeff-dean-leaving-company-after-27-years.html">现金流首次出现负值</a>。谷歌目前正<a href="https://www.bloomberg.com/news/articles/2026-08-06/google-shifts-ai-power-to-california-in-race-against-anthropic-openai">将AI领导层集中到加利福尼亚</a>。</p> <p><strong>该公司正在转变其AI科学战略</strong><br>它正从专用工具（如DeepMind的AlphaFold）转向能够更自主开展研究的智能体AI系统。<a href="https://www.technologyreview.com/2026/05/22/1137813/google-i-o-showed-how-the-path-for-ai-science-is-shifting/?utm_source=the_download&utm_medium=email&utm_campaign=the_download.unpaid.engagement&utm_term=*%7CSUBCLASS%7C*&utm_content=*%7CDATE:m-d-Y%7C*">在我们的近期报道中了解更多</a>。<br></p> <hr class="wp-block-separator has-alpha-channel-opacity is-style-default" style="margin-top:0;margin-bottom:0"/> <p></p> <p>我们很高兴地宣布，<em>MIT Technology Review</em>现已登陆<a href="https://www.instagram.com/technologyreview/reels/">Instagram Reels</a>和<a href="https://www.youtube.com/@technologyreview/shorts">YouTube Shorts</a>，以及<a href="https://www.linkedin.com/company/mit-technology-review/">LinkedIn</a>和<a href="https://whatsapp.com/channel/0029VaYJXzgChq6Dm1g6od0L">WhatsApp</a>。您可以在自己喜欢的渠道和格式中直接听到<em>MIT Technology Review</em>记者的声音，他们帮助您理解瞬息万变的科技世界下一步将发生什么，以及这对您意味着什么，同时还能让您一窥他们的报道幕后。<br><br>为了让您快速上手，请让我们的资深气候记者Casey Crownhart <a href="https://youtube.com/shorts/14UBFdNkQOo?feature=share">为您介绍激光如何帮助为核能提供燃料</a>，或者让资深调查记者Eileen Guo <a href="https://www.instagram.com/p/DXr83EzCVF-/">帮助您了解World新推出的身份验证举措对我们的隐私意味着什么</a>。</p> <p class="has-medium-font-size"><strong>必读文章</strong></p> <p><em>我在互联网上为你搜罗了今天最有趣/最重要/最惊悚/最迷人的科技故事。</em></p> <p><strong>1 Meta成为最新一家声称其AI入侵了另一家公司的公司</strong><br>它将责任归咎于独立网络安全测试员的“配置错误”。(<a href="https://edition.cnn.com/2026/08/05/tech/meta-ai-hacking">CNN</a>)<br><em>+ 据悉，涉及的模型是Muse Spark 1.1。</em>(<a href="https://www.theinformation.com/articles/meta-ai-model-hacked-another-company-cybersecurity-testing">The Information</a> $)<br><em>+ 此前OpenAI和Anthropic的模型也发生过类似入侵事件。</em>(<a href="https://www.theguardian.com/technology/2026/aug/05/meta-ai-model-hack-training">Guardian</a>)<br><em>+ 这就是AI代理为实现目标可能撒谎的原因。</em>(<a href="https://www.technologyreview.com/2026/08/03/1141009/heres-why-ai-agents-lie-and-cheat-to-reach-their-goals/?utm_source=the_download&utm_medium=email&utm_campaign=the_download.unpaid.engagement&utm_term=*%7CSUBCLASS%7C*&utm_content=*%7CDATE:m-d-Y%7C*">MIT Technology Review</a>)<br><br><strong>2 三星和SK海力士在美国出口管制背景下测试中国芯片设备</strong><br>这两家韩国芯片巨头正在对冲更严格限制的风险。(<a href="https://www.reuters.com/world/china/samsung-sk-hynix-test-chinese-chip-tools-hedge-against-us-risks-2026-08-05/">Reuters</a> $)<br><em>+ 北京已对Palo Alto Networks展开调查。</em>(<a href="https://www.bloomberg.com/news/articles/2026-08-06/china-launches-cybersecurity-review-into-palo-alto-networks?srnd=phx-technology">Bloomberg</a> $)<br><em>+ 在习近平与特朗普峰会前夕，美中科技紧张局势不断升级。</em>(<a href="https://www.scmp.com/economy/policy/article/3363062/disputes-over-ai-robotics-trade-mount-ahead-planned-xi-trump-summit-us?utm_source=feedly_feed">SCMP</a>)<br> <br><strong>3 伦敦刚刚批准机器人出租车获得运营许可</strong><br>条件是目前仍需配备人类司机。(<a href="https://www.bbc.co.uk/news/articles/c3w07qg43j6o">BBC</a>)<br><em>+ Uber计划投入超过100亿美元扩展其机器人出租车网络。</em>(<a href="https://www.ft.com/content/04d6fcb4-03e4-4ee5-9017-ccaa7808c934">FT</a> $)<br> <br><strong>4 科学家培育出不会引发过敏的基因编辑犬</strong><br>他们使用CRISPR移除了导致过敏反应的蛋白质。(<a href="https://www.wired.com/story/gene-edited-puppies-wont-trigger-allergies/">Wired</a> $)<br><em>+ 目前他们希望获得批准，开始销售这些比格犬。</em>(<a href="https://www.newscientist.com/article/2583502-two-dogs-lack-the-protein-that-causes-allergies-thanks-to-gene-editing/?utm_campaign=RSS%7CNSNS&utm_content=home&utm_medium=RSS&utm_source=NSNS">New Scientist</a> $)<br><em>+ 其他公司也在计划培育基因编辑婴儿。</em>(<a href="https://www.technologyreview.com/2025/10/31/1127461/heres-the-latest-company-planning-for-gene-edited-babies/?utm_source=the_download&utm_medium=email&utm_campaign=the_download.unpaid.engagement&utm_term=*%7CSUBCLASS%7C*&utm_content=*%7CDATE:m-d-Y%7C*">MIT Technology Review</a>)<br> <br><strong>5 OpenAI已请求法官驳回苹果的商业机密诉讼</strong><br>这家ChatGPT开发商称苹果的指控“毫无依据”。(<a href="https://www.theverge.com/tech/976042/openai-apple-trade-secrets-lawsuit-dismissal-request">Verge</a>)<br><em>+ 并称该诉讼旨在阻止员工外流。</em>(<a href="https://www.ft.com/content/d9e9cbbe-8443-4c10-821a-1834e4965236?syn-25">FT</a> $)</p>