亚马逊放弃其OpenAI电影项目的原因、数据中心员工发起抵抗，以及Meta泄露员工数据
内容总结：
Uncanny Valley播客本周热点：AI与好莱坞的纠葛、数据中心争议及科技巨头内幕
本周，WIRED旗下播客《Uncanny Valley》的主持人围绕多个热点话题展开讨论。首先，亚马逊旗下米高梅影业突然叫停了一部关于OpenAI CEO萨姆·奥特曼的传记电影《Artificial》。该片由著名导演卢卡·瓜达尼诺执导，主演安德鲁·加菲尔德饰演奥特曼。据称影片对奥特曼的描绘并不友好，而亚马逊作为OpenAI的投资方之一，此举被外界解读为有意为奥特曼“保驾护航”。
与此同时，谷歌DeepMind向独立电影公司A24投资7500万美元，用于开发AI电影制作工具。主持人指出，虽然AI在电影行业的实际应用仍相对有限（主要在故事板绘制、转描等劳动密集型环节），但科技巨头与影视产业的深度捆绑已经成为不可忽视的趋势。
在数据中心话题上，节目探讨了日益加剧的社会反弹。从电工到软件工程师，越来越多的从业人员开始抵制数据中心建设。调查显示，美国超过40%的家庭现居住在距数据中心5英里范围内。电力成本上升、水资源消耗和噪音污染是居民抗议的主要原因。更有意思的是，甚至连参与建设的电工都抱怨，因为从事数据中心工作而在社交场合“谈对象都变难了”。
Meta方面，该公司推出了一项颇具争议的员工监控计划——在员工设备上安装软件，追踪每一次键盘敲击和屏幕活动，以此训练AI。但本周WIRED独家披露，Meta意外将这些敏感数据在公司内部暴露，导致任何员工都能看到同事的屏幕活动。Meta随后宣布暂停该数据收集计划。
最后，节目还关注了Anthropic与特朗普政府关系的最新进展。此前因NSA发现其AI模型存在安全漏洞，Anthropic被迫撤回两款先进模型。近期双方重启谈判，关键在于Anthropic创始人兼CEO达里奥·阿莫代伊不再参与对话——一位政府官员直言：“汤姆·布朗（另一位创始人）不像达里奥那样‘古怪’，他能够正常沟通。”
中文翻译：
本周在《Uncanny Valley》节目中，主持人将讨论亚马逊引起争议的决定——放弃卢卡·瓜达尼诺执导、关于OpenAI的萨姆·奥尔特曼的电影，据报道该片对其形象描绘并不正面。与此同时，谷歌DeepMind与独立电影工作室A24达成7500万美元的全新合作，AI对我们的电影产业究竟造成了多大影响？主持人还将深入探讨近期从电工到软件工程师等工人群体反对数据中心的浪潮。此外：Meta的员工数据追踪项目在发生大规模泄露后被暂停，而Anthropic因CEO达里奥·阿莫代不再参与会议，与政府的关系正在改善。
本期节目提到的文章：
- 《A24知道你对他们与谷歌AI合作很恼火》
- 《一些电工认为建设数据中心是种背叛》
- 《Meta在内部数据泄露后暂停员工追踪项目》
- 《特朗普白宫对Anthropic CEO达里奥·阿莫代感到厌倦》
您可以在Bluesky上关注布莱恩·巴雷特（@brbarrett）、佐伊·希弗（@zoeschiffer）和莉亚·费格（@leahfeiger）。来信请发送至[email protected]。
收听方式
您始终可以通过本页面的音频播放器收听本周播客，但如果您想免费订阅以获取每期节目，请按以下步骤操作：
如果您使用iPhone或iPad，请打开名为“播客”的应用，或直接点击此链接。您也可以下载Overcast或Pocket Casts等应用并搜索“uncanny valley”。我们在Spotify上也有账号。
文字记录
注意：此为自动生成文字记录，可能存在错误。
布莱恩·巴雷特：嗨，我是布莱恩。在开始之前，先说两件小事。如果您一直喜欢收听我们的节目，我们希望您能花点时间在您常用的播客应用中给我们评分，这确实能帮助我们触达更多听众。第二，如果您有任何关于科技、隐私或政治方面的问题，希望我和佐伊、莉亚来解答，现在正是时候，请将问题提交至[email protected]。无论问题大小，我们都想倾听您的声音并为您找到答案。好了，节目开始。
佐伊·希弗：欢迎收听《连线》杂志的《Uncanny Valley》。我是佐伊·希弗，商业与产业总监。
布莱恩·巴雷特：我是布莱恩·巴雷特，执行编辑。
莉亚·费格：我是莉亚·费格，政治与科学总监。
佐伊·希弗：今天节目的话题是：AI与好莱坞。我们将讨论亚马逊米高梅工作室在制作接近尾声时突然决定放弃OpenAI传记电影。与此同时，谷歌DeepMind正投资7500万美元与电影工作室A24共同开发AI工具。AI与电影产业正变得越来越紧密，我们准备探讨这一切将走向何方。
莉亚·费格：我们还会深入探讨近期围绕数据中心的一些争议。随着国家和地方对数据中心建设的反对声浪日益高涨，一些电工甚至拒绝参与建设，而且他们并非孤例。一群亚马逊员工声称，他们因公开支持监管而正在接受调查。
布莱恩·巴雷特：如果您最近几周一直在收听我们的节目或阅读《连线》，您就知道我们一直在密切报道Meta内部正在发生的危机。本周我们又有一条独家新闻：在公司内部泄露了该系统的敏感数据后，那个追踪员工每一次击键和屏幕活动的争议性系统已被暂停。我们将讨论这一系列令人沮丧的事件是否真能促使公司内部发生改变。
莉亚·费格：在节目的后半段，我们还将带来最新消息：既然CEO达里奥·阿莫代不再参加会议，Anthropic与政府之间的谈判似乎正在改善。
佐伊·希弗：好的，各位，我真的很兴奋要聊聊这部电影《Artificial》的商业故事，它突然被亚马逊米高梅工作室放弃了。先给大家梳理一下背景。《Artificial》是《请以你的名字呼唤我》和《挑战者》的导演（两部都是我觉得很出色的电影）执导的，是一部关于OpenAI的传记剧情片，聚焦于“那场风波”——2023年11月，萨姆·奥尔特曼被董事会突然解雇，然后在几乎整个公司反抗后又迅速被重新聘用。我实在不敢相信他们居然拍了整部电影。
布莱恩·巴雷特：是啊。
佐伊·希弗：它被描述为“AI时代的《社交网络》”。演员阵容星光熠熠，安德鲁·加菲尔德饰演萨姆·奥尔特曼，莫妮卡·巴巴罗饰演前OpenAI CTO米拉·穆拉蒂。电影算是中等预算，据称制作成本花了4000万美元，但基本上已经快完成了，这时亚马逊宣布放弃这部电影，称“由其他工作室发行会更合适”。这一决定招致批评，因为这被视为亚马逊在帮萨姆·奥尔特曼的忙，而他在电影中的形象相当负面。我很想知道你们的看法。
莉亚·费格：我立刻就有很多想法。第一，安德鲁·加菲尔德，我们早就猜到了。我们一直都知道会是他演。扮演爱抱怨的科技宅男绝对在他的拿手戏范围内。
佐伊·希弗：他曾在《社交网络》中饰演被马克·扎克伯格排挤出去的Facebook联合创始人，很出名。
布莱恩·巴雷特：我觉得这里还有一层背景：亚马逊在OpenAI投资了500亿美元，对吧？所以，当他们说“这部电影由其他工作室发行会更合适”时，我认为他们真正的意思是“这个工作室由其他电影来运作会更合适”，对吧？这真的像是——
莉亚·费格：确实如此。
布莱恩·巴雷特：——在保持距离。这件事最让我触动的是我们一直知道会发生、也正在发生的事情，但电影产业和科技产业目前相互交织的程度之深。亚马逊拥有米高梅。派拉蒙正被埃里森家族收购，拉里·埃里森，显然是甲骨文的创始人。所以所有这些科技亿万富翁现在都或多或少地完全融入了电影产业，这真的将决定哪些电影能被制作出来，哪些不能。而这就是一个非常显著的例子。
佐伊·希弗：是啊。我的意思是，我可以亲身证实，有很多关于Twitter/埃隆·马斯克的电影项目都在洽谈中，但由于显而易见的原因，很难真正搬上银幕。我确实认为，根据我们听到的所有消息，这部电影对萨姆·奥尔特曼是不利的。事实上，“那场风波”本身就对萨姆·奥尔特曼很不利，因为他的许多高管之所以反叛他并策划了所谓的“政变”，是因为他们认为他表里不一、说谎，对不同的人说不同的话，只为了说对方想听的。在电影中，看起来OpenAI的前首席科学家伊利亚·苏茨克弗真的成了英雄。他后来创立了另一家专注于安全人工智能的公司。所以，这对亚马逊来说是一个复杂的项目，我并不完全感到意外。除了你刚才提到的投资，布莱恩，他们最近还达成了一项价值380亿美元的算力交易。
布莱恩·巴雷特：而且萨姆·奥尔特曼去年还是杰夫·贝佐斯婚礼的座上宾。这里有人情、有金钱，各种因素交织在一起。
佐伊·希弗：没错。
布莱恩·巴雷特：你多少会希望，理想情况下，对吧，就像他们对待媒体产业那样——当这些亿万富翁接管这些公司时，采取一种不干涉的态度：你们做你们的事，我相信艺术。我这么做是因为我相信它。但现实并非如此。在媒体产业，这种情况也越来越少见了。杰夫·贝佐斯显然在《华盛顿邮报》做出了一些相当大的改变，这是一种施加影响力的做法，在理想情况下是不应该发生的。有权有势的利益集团拥有好莱坞公司已经很久了，但这确实是一个如此明目张胆的例子。
佐伊·希弗：是的。而且我认为对OpenAI来说，他们现在对公众舆论非常敏感。他们觉得——而且他们不是唯一这么觉得的，Anthropic也有同感——AI越来越不受欢迎了。已经有一些关于萨姆·奥尔特曼在OpenAI上市前再次被赶下台的讨论，可能就在今年，最晚明年。所以，虽然你可能会说，“哦，他们可以不在意这个，他们专注于其他事情”，他们已经习惯了负面的媒体报道，也许还有负面的形象刻画。但我认为他们正越来越努力地控制信息。
布莱恩·巴雷特：而且还有媒介，对吧？我认为本周发生的另一件引人注目的交易是谷歌DeepMind宣布投资7500万美元给备受喜爱的独立电影宠儿A24，用于创建AI工具。莉亚，我知道你对这件事有强烈的、而且是很遗憾的感受。
莉亚·费格：是啊。但问题是，我有点讨厌A24。
布莱恩·巴雷特：哦。
莉亚·费格：它已经够程式化的了。我心想，“啊，又来一部这种特定棕褐色调、用这些特定演员演这些特定角色的电影。”当我看到A24与谷歌DeepMind合作的新闻时，我就想，“完美。这本来就是一个等待发生的算法。他们当然会把它交给AI了。”
佐伊·希弗：好吧，这个内容就是为我量身定做的。我是A24的超级粉丝。不过，我和好莱坞的很多业内人士聊过这件事，因为虽然有这么多合作，但我们看到OpenAI和Anthropic这些大型AI实验室暗示很快就能用AI制作电影。然后你真的看到一部AI生成的电影时，你会想，“肯定不是这样。”但我听到的是，AI正越来越多地被用于非常特定的目的。比如故事板绘制。我相信在新闻稿中，谷歌DeepMind和A24也提到了这可能是用途之一。还有像转描这样的工作，基本上是电影制作过程中非常劳动密集型、特定的领域，过去需要大量人力，成本很高，而现在确实可以自动化了。在一些人看来，这些是AI非常好的应用场景。我认为我们会开始在银幕上看到越来越多由AI生成的特定镜头。但如果看到一部完全由AI生成、从头到尾在大银幕上映的长片，而且不是一次性的噱头，我会非常惊讶。当意识到自己接触的东西是AI生成而自己却不知道时，会产生一种“恶心”的感觉，我认为这种感受不容忽视。另外还有一点是，它能否真正创作出质量足够高、能出现在那种语境下的作品？
布莱恩·巴雷特：我也确实认为，有一点，这也呼应了你之前的观点，佐伊，当人们看到一项7500万美元的交易时，会假设这是用来训练AI的。会假设，“哦，现在谷歌要把A24的所有东西都吸收进那个‘博格’里”——但并非如此。所以有趣的是，这显然仍有争议，但它确实是用于那些——试图找到那些专门构建的工具来支持非AI电影的创作，这一点被忽略了，但确实如此。
佐伊·希弗：你需要真人。我认为之前有人担心谷歌会用它来训练模型，但协议中明确排除了这一点。我猜，如果让我猜的话，谷歌大概想要获取那些IP，但A24如果给了他们那种权限，那将是非常糟糕的品牌举动。
布莱恩·巴雷特：我们谈过万事万物如何相连——再给这关联加一条：A24的投资者之一是Thrive Capital。这是一家由贾里德·库什纳的兄弟乔什·库什纳拥有的风险投资公司。Thrive Capital将OpenAI作为一项主要投资，持有SpaceX的大量股份，而华纳兄弟即将被埃里森家族收购——真是关系错综复杂，每个人都与其他人有牵连。
莉亚·费格：电影产业试图与AI亲近。也有其他一些人因为各种不同的原因讨厌AI。除了对创造力的担忧，还有一个我们一直在关注的现象：全美各地对数据中心持续且日益增长的抵制。根据皮尤研究中心的分析，如今美国超过40%的家庭距离正在运营的数据中心在5英里范围内。随着大型AI公司投入数十亿美元用于基础设施建设以应对需求，这股建设热潮仍在继续。《连线》报道过居民们反对在他们社区建设数据中心的种种原因，从导致电费上涨、水资源短缺到噪音等等，不一而足。而理论上应该是这些项目的实施者，工人们实际上也在尽自己所能进行抵制。你们一直在关注这件事吗？
布莱恩·巴雷特：是的，看到这种演变很有意思。我认为我们已经讨论了很多关于社区抵制的问题，这仍然是一个重要的故事，但我们在《连线》上也有几篇报道，其他地方也有关于直接相关人员的报道。《连线》的职员记者卡罗琳·哈斯金斯写了一篇很棒的文章，关于电工——你知道吗？电工对建设数据中心至关重要。他们中的一些人现在开始说，“等等，这算不算是一种背叛？在一个数据中心工作是否意味着你在某种程度上违背了更广泛的原则，不仅仅是电工的原则，而是做人的原则？”
佐伊·希弗：你错过了那个故事最精彩的部分，就是那个电工说，一旦告诉别人他是建设数据中心的电工，约会就变得非常困难。
布莱恩·巴雷特：就像Cybertruck一样。就像拥有一辆Cybertruck。
佐伊·希弗：确实让我想起了那个。
布莱恩·巴雷特：在数据中心工作就是新的拥有一辆Cybertruck。
佐伊·希弗：我认为困难之处在于，嗯，是的。人们不想住在数据中心附近。它们会抬高能源账单，又吵又亮，还占用各种资源。我认为这非常合理。就全国性的经济举措而言，数据中心在某些方面感觉像是我们仅有的东西。我这么说会被批判吗？我只是觉得，这是大事，AI是美国正在押注的大事，我认为试图减缓或逆转这一趋势，而不是试图推动它们变得不那么令人讨厌，是很复杂的。
布莱恩·巴雷特：但我认为事情是这样的。我同意。建设数据中心和投资AI在目前阶段就是美国经济。一旦这停止了——
佐伊·希弗：没错。
布莱恩·巴雷特：那将会非常非常痛苦。不过我认为很多抵制的理由是，数据中心来了其实不会给你当地带来太多直接的经济利益。通常，比如“哦，铁路要通到我们镇了”，你至少还能从铁路那里获得额外的商业机会，对吧？
佐伊·希弗：对。
布莱恩·巴雷特：情况就是这样。
佐伊·希弗：哦，天哪。
布莱恩·巴雷特：我老到经历过那种时代。
佐伊·希弗：这是我唯一能接受的关于铁路的例子。
布莱恩·巴雷特：谢谢。是的，很好。我做到了。所以我认为这是原因之一。与此同时，这些公司内部也有人带头反对。所以不仅仅是直接受影响的人，还有那些实际上应该从这些事情中受益的人，比如电工，比如亚马逊内部的人。例如，一些亚马逊员工最近敦促西雅图市议会监管数据中心。所以，即使是那些可能从这些事情中获得经济利益的人也在抵制。我明白你的意思，佐伊，但我认为有一定的——对我来说，人们说“等等，等一下”是有道理的，因为这并不是每个人都能参与的经济。
佐伊·希弗：不，完全正确。我想说清楚，我完全理解这种抵制。如果在我家附近建一个，我也会反对。我认为不想住在它们附近是完全合理的。我只是觉得，当我们的经济如此依赖这个产业时，情况就很复杂了。这不是不监管的理由，但我认为要逆转它将非常复杂。我也一直很好奇，莉亚，我想听听你的看法，就是看看哪些政客真的——我有点惊讶伯尼·桑德斯成了数据中心的强烈反对者。
莉亚·费格：是的。参议员伯尼·桑德斯和众议员亚历山德里娅·奥卡西奥-科尔特斯（AOC）提出了《数据中心暂停法案》，该法案将暂停新建AI数据中心，直到有真正的国家保障措施出台。这在很多方面听起来像一个非常典型的左翼议题，但令人惊讶的是，这实际上是两党都支持的。来自各政治派别的人都参与其中，因为他们的选民在联系他们并质问：“你以我的名义在做什么？你在我的后院做什么？这对我有什么好处？这对我有什么害处？”所以，我认为最让我震惊的是这个问题上的两党合作。
佐伊·希弗：这很有意思，因为我觉得——当然这纯粹是我的猜测——但基于OpenAI如何谈论数据中心，在特朗普政府上台的第一天就高调出面，大力倡导大型数据中心建设项目。我看到克里斯·莱汉，公司的首席全球事务官，之前在爱彼迎担任要职，再之前是个政治操盘手，我想这可能是他和OpenAI认为对公司非常有利的事情。这有点像“美国优先，建吧，宝贝，建吧”这样的信息。
莉亚·费格：我们给所有人提供工作。
佐伊·希弗：他们只是误判了形势。他们没有意识到这个问题会变得多么棘手。现在，当他们每次有新数据中心都发布新闻稿之后，要改变立场非常困难。现在变成了，“哎呀，我们得保持低调，因为人们真的很不喜欢这个。”
布莱恩·巴雷特：而且如你所说，他们无论如何也无法逆转，除非你能把数据中心建到太空里去，因为他们需要算力。
佐伊·希弗：顺便说一句，这将非常困难。
布莱恩·巴雷特：如果不是不可能的话。佐伊，有没有可能，那种内部的异议，对吧，电工说“我不这么认为”，公司内部的员工说“嘿，我们也不喜欢数据中心”，有没有可能这会在任何方面改变这些建设、这些公司、这些支出的轨迹？
佐伊·希弗：我会非常惊讶。我不想说绝对不可能，因为我们见过例子，著名的谷歌员工曾团结一致，反对“Maven项目”以及针对中国的一些审查搜索项目，并确实让这些项目暂停了启动。
布莱恩·巴雷特：快速补充一下，“Maven项目”基本上是跟五角大楼合作，对吧，使用谷歌技术为国防部服务？
佐伊·希弗：没错，正是。所以，是的，以前发生过。以后也可能再次发生。我想说两点。第一，我们看到的来自小时工的反抗，在整个劳动力队伍中是微不足道的。他们正在招募成千上万人。我听说他们支付的薪水比通常这类工作要高得多。所以我认为，对于一个历史上需要大量工作的行业来说，总会有人愿意从事这些项目，然后我们会听到零星的反抗和抵制的声音，这当然有新闻价值，也相关，但不是没有人干，但我仍然认为他们能够招募成千上万的人。我还要说，在公司层面，虽然我们开始看到更多来自企业员工的抵制，他们更公开地反对自己公司的所作所为，但这仍然远低于2018年左右的水平。
布莱恩·巴雷特：特别是现在这些工程师的就业市场与当时完全不同了。而且我认为几年前也有一个转折点，谷歌解雇了一群抗议的员工，而之前他们或多或少是允许这类事情发生的。我要说，有一件事能让管理层倾听员工的声音，那就是一个巨大的失误，就像我们最近在Meta看到的那样。我们之前讨论过，Meta在员工设备上安装了软件，用来追踪每一次击键、所有屏幕活动，基本上是对员工进行完全监控，以帮助训练AI。本周，《连线》爆料称，该公司显然将这些会话中的潜在敏感信息暴露在Meta内部，任何人都可以访问。所以，如果你是Meta员工，你在屏幕前所做的一切，你的任何同事都可能看到。你能想象吗？
佐伊·希弗：不能，而且我觉得我们得再稍微回溯一下，为听众理清前因后果。所以，马克·扎克伯格觉得自己在AI竞赛中落后了。他投入数十亿美元建立这个新的AI实验室，花大价钱招募特定人才，给他们大量资源来追赶并构建前沿模型。然后公司说，“我们在AI建设上花了太多钱，所以实际上不得不裁掉你们中的很多人，裁掉10%的员工。”然后他们说——
布莱恩·巴雷特：8000人。
佐伊·希弗：对，8000人。然后他们说，“顺便说一句，你的笔记本电脑上将被安装监控技术，追踪你的击键，不管你喜不喜欢，目的是为了训练那个刚刚引发这些裁员的、花哨的AI实验室试图构建的模型。然后，哦，顺便说一句，我们不小心把这些数据全部暴露了。”
布莱恩·巴雷特：“但我们几个月后会举办一个黑客马拉松来提升士气。”这件事可能带来的一线希望或者说是好事是，Meta确实宣布暂停数据收集项目，同时调查到底发生了什么，但这只是暂停。只是暂停。
佐伊·希弗：他们确实说了“暂停”。然后那天晚上我们更新了那篇报道，他们非常明确——他们没有在公司范围内说过泄露这件事。我们是从很多不同的消息来源那里听说的。他们仍然对此守口如瓶。然后深夜，我们终于得到了他们发出的消息，他们非常明确地说：“我们将评估这里发生了什么，确保在重新启动这些工具之前不会再发生，”这就是传达的信息。
布莱恩·巴雷特：很多道歉的“代言人”是Meta的CTO安德鲁·博斯沃思，绰号“Boz”。这让事情更添了几分讽刺，因为在这件事发生、在这起内部泄露事件发生之前，人们曾直截了当地问他：“从隐私角度来看，这难道没有潜在危险吗？”他告诉员工：“实际上，它受到严格控制，我们使用了与其他敏感数据集相同的保护标准、存储系统和访问控制。”这让你不禁担心Meta掌握的其他敏感数据集——
佐伊·希弗：确实。
布莱恩·巴雷特：会怎样。
佐伊·希弗：完全同意。我很期待我们安排的那个专题，标题叫“Boz和他的Meta伙伴们”。我觉得布莱恩，我不知道，只是感觉这是一个你会说“好”的故事。
布莱恩·巴雷特：没错。再怎么强调也不为过，Meta内部互相称呼为“Metamates”。你们想听一段泄露的音频吗？上个月马克·扎克伯格与员工讨论培训计划，解释为什么他认为这至关重要。
莉亚·费格：当然想听。
佐伊·希弗：绝对。
马克·扎克伯格（存档音频）：我们正处于一个阶段，基本上AI模型通过观察非常聪明的人做事来学习。总的来说，这家公司员工的平均智力水平，远远高于你能找来执行任务的普通人群的平均水平。所以，如果我们想教模型写代码，比如说，那么让内部员工构建工具或解决任务来帮助教模型如何写代码，我们认为这将极大地提高我们模型的编码能力，速度会快于行业内其他公司，因为他们公司里没有成千上万的顶级工程师。
佐伊·希弗：你真的很擅长做会议记录，你太擅长了。所以我们不得不把这份工作给你。
布莱恩·巴雷特：你们都会非常擅长用AI替代自己，以至于我们会让你们做得更快。
莉亚·费格：我们能聊聊另一家AI公司吗，各位？
佐伊·希弗：哦，天哪，莉亚。
莉亚·费格：你的梦想，佐伊，你做梦都想——
佐伊·希弗：你走开——
莉亚·费格：听我这么说。
佐伊·希弗：消失了一段时间，你就准备好没完没了地聊AI了。
布莱恩·巴雷特：我们中的一员。我们中的一员。
莉亚·费格：嗯，这事跟Anthropic和美国政府有关。如我们所知，Anthropic与特朗普政府就他们最先进的模型Fable 5和Mythos 5发生了一点摩擦，原因是美国国家安全局确认存在越狱和禁用这些模型安全护栏的方法。此后，双方都在努力寻找前进的道路。有时看似有进展，有时则不然。但最近，《连线》的雨果·洛厄尔在他的通讯《Inner Loop》中报道称，政府在最近几天与Anthropic进行了多次电话交谈，似乎正在采取一些行动，一些令人鼓舞的行动，但这主要归功于他们不再需要与该公司的CEO达里奥·阿莫代打交道了。各位，这句话我将永远铭记，因为政府非常高兴Anthropic的联合创始人汤姆·布朗和公共政策主管莎拉·赫克一直在主导对外联络。我们有一位官员对此的引述是这样的。我永远忘不了。“汤姆·布朗不像达里奥那样古怪，他能够进行真正的交流。”
佐伊·希弗：难以置信。我的意思是，从来没有人说过达里奥不古怪。那家伙确实是个怪人。
布莱恩·巴雷特：我喜欢这个引用，对我们来说。我为他感到非常难过。
莉亚·费格：我为他感到非常难过，但这也恰恰说明了——这里发生了什么。不，不，不是技术问题变了。而是跟谁打交道变了——
佐伊·希弗：没错。
莉亚·费格：这，我不知道怎么说，也许这就是特朗普政府的缩影。这些对话一直在高层工作组层面、技术人员层面进行。
佐伊·希弗：我觉得我们需要快速梳理一下利害关系，给那些没有一直关注这出戏每一个曲折情节的人——哇，你的大脑一定非常健康，我很羡慕，但——
布莱恩·巴雷特：那一定很好。是啊。
佐伊·希弗：那一定很好。美国政府，在国家安全局基本上断言存在越狱模型的方法之后，对这些模型实施了出口管制，并基本上表示Anthropic不允许外国公民访问这些模型。Anthropic没有技术手段来做到这一点。所以他们不得不做的是，取消所有人对这两个模型的访问权限。这对公司来说是一件非常大的事。所以这些对话基本上是在试图——从Anthropic方面来说——弄清楚他们需要做些什么才能让这些模型重新投入市场。而从美国政府方面来说，则是需要Anthropic做些什么才能让他们感到放心，这些模型可以发布，而不会成为巨大的网络安全噩梦？
布莱恩·巴雷特：嗯，特别是，根据我们的了解，似乎特朗普政府要求的东西基本上是不可能的。他们要求一个无法被越狱的模型，这根本不存在。这些模型的本质是，你最多能做到的——Anthropic已经说过“嘿，我们能行”——就是让它几乎不可能实现通用越狱，通过通用越狱你可以让模型做任何你想做的事。但你可以想象，在像这样的对话中，尤其是有一个人能够从非技术层面和政治层面来讨论这些问题，而且也许能够压制住自己关于AI的那些高深想法，你可以看到这在这样的情况下有多么重要，并且对所有公司来说都将继续重要。他们迟早都要到白宫去，为为什么这个超级强大的模型应该被发布而陈述理由。
佐伊·希弗：你真的需要一个人能不用“费米悖论”、“正交性”这样的词就解释清楚。他们得能说人话。
莉亚·费格：所以事情的最终结局会是这样吗？国家安全方面的担忧，会因为有人真的认真听了他们的公关人员关于如何在公众面前讲话的建议超过三分钟就能缓解？
布莱恩·巴雷特：我的意思是，是的。我甚至可以说得更远一点。如果他们两个月前就听了公关人员的建议，或者他们的公关人员采取了不同的策略，我想我们一开始就不会走到这一步。
佐伊·希弗：是啊。我的意思是，但这就是有趣的地方，因为Anthropic花了大量时间公开宣称：“AI非常危险。AI可能会终结人类。我们的模型非常可怕，在网络安全和黑客方面非常强大，以至于我们不能公开发布它们。”所以我的确觉得他们，尤其是达里奥·阿莫代，将自己和公司定位在了这个非常特定的位置上，实际上还同时说我们需要大量的政府监管。当他们说所有这些话，而他们与五角大楼又没有牢固的关系时，他们是在走钢丝，非常危险的一步。不久之前，他们还因为试图划清界限而闹得很不愉快，说：“你们不能把我们用于任何你们想要的军事行动。你们不能用它们来制造自主武器”等等。而五角大楼则说：“你们不能决定我们如何处置这些模型。这是美国的AI，我们想做我们想做的事。”
布莱恩·巴雷特：广告之后，我们将分享本周的WIRED/TIRED推荐。请继续收听。
佐伊·希弗：到了我们的WIRED/TIRED环节。你们懂的，新鲜酷炫的东西是WIRED，过时乏味的东西是TIRED。
莉亚·费格：好的。我先来个积极的。我的WIRED是卡罗琳·卡洛威。她——
布莱恩·巴雷特：哇。
莉亚·费格：回到了我们的信息流中，回到了我们的信息流中，各位，正在经历一场糟糕的分手，并且通过发帖来应对。
布莱恩·巴雷特：用两句话跟听众介绍一下卡罗琳·卡洛威，因为我在过去十年上网太多了，但不是每个人都这样。所以简单说一下她是谁，为什么这这么让人兴奋。
莉亚·费格：卡罗琳·卡洛威有点像曾经的骗子，现在是市中心Dime Square的女孩。我鼓励你，如果你不知道这个女人是谁，去探索一下这个兔子洞吧。非常非常有趣。强烈推荐。
佐伊·希弗：所以她是WIRED。
莉亚·费格：是的，她是我的WIRED，而我的TIRED是《Feed Me》的艾米丽·桑德伯格。
布莱恩·巴雷特：莉亚一上来就这么猛。
佐伊·希弗：她在开火。
布莱恩·巴雷特：开火了。
莉亚·费格：我在开火。佐伊知道我已经有这个苗头一段时间了。艾米丽·桑德伯格的Substack曾经是我最喜欢的Substack之一。我认为她当时在做一些非常酷的事情。她把商业新闻、科技新闻和文化新闻融合在一起。最近，在过去几周、几个月里，很多人也确实开始关注艾米丽·桑德伯格。她得到了很多媒体报道和合作机会，我很难用完全恰当和友善的方式来表达，但不管怎样，她本周最新的通讯采访了Meta的马克·扎克伯格，聊他的时尚风格和哈佛时光，这让我尴尬得要死。所以，TIRED，艾米丽·桑德伯格。很抱歉。我认为你还能扭转局面。我真的这么认为，但你写的东西现在都很尴尬，我受够了。
佐伊·希弗：好的。这只是莉亚的个人观点，艾米丽。我仍然喜欢你的通讯。
布莱恩·巴雷特：莉亚的观点并不代表整个《Uncanny Valley》节目组。
佐伊·希弗：不，实际上我们分歧很大。我们在播客过程中意识到了这一点。好了，布莱恩，我们来说说——
布莱恩·巴雷特：不，我接不了那个。
佐伊·希弗：我希望你有一个最温和的WIRED/TIRED。
布莱恩·巴雷特：我接不了那个。不。我的WIRED是交朋友，TIRED是保持……不，跟那个比起来，我的太无聊了，但我还是说吧。莉亚，你该后说的。你应该坚持的。我的WIRED是便宜的电动卡车。便宜的电动卡车。我说出来了。别那样看着我。Slate是一款新的电动卡车，已经研发了一段时间了。它之前有很多宣传，但他们将以25,000美元的价格出售，这仍然是一大笔钱。但美国新车的平均价格，有人知道是多少吗？50,000美元是新车的平均价格。所以他们推出了一款电动卡车。全是灰色的，非常朴素。没有任何花哨装饰，但价格是25,000美元。所以这是一款任何人都能买得起的电动车。他们能做到这一点是因为这辆车什么配置都没有。车窗得手摇，每样东西都得加钱——这就像是电动车里的精神航空，你什么都得付钱——
佐伊·希弗：等等，这听起来是TIRED。我们不想要这个。
布莱恩·巴雷特：不，这是WIRED。我喜欢这个。我认为这是WIRED，有一款你买得起的电动车，然后你可以添加各种配件，有超过200种配件可以加。所以如果你想要车上有其他东西，比如说想要车有颜色，或者想要自动车窗之类的，你可以一件一件地加，每辆车都可以不同，而不是你一进店就得花50,000美元买一堆你可能不想要或不需要的东西。我认为如今，尤其是在政府退坡对电动车的激励、大力推动石油的背景下，能看到一些人们真正能买得起、能真正定制、续航里程还不错的车，真的很好，很出乎意料。这就是我的WIRED。TIRED是与之相反的任何东西。
佐伊·希弗：我就知道当你说精神航空的时候，那个听起来是负面的。我真的很想让我的下一辆车是Rivian。我全力支持电动。是的。
布莱恩·巴雷特：我在等R3，但显然要到2030年才会出来。
佐伊·希弗：哦，天哪。是的。
布莱恩·巴雷特：坚持住。
莉亚·费格：我就想要斯巴鲁。我想要一辆斯巴鲁电动车。我把这个愿望说出来。
布莱恩·巴雷特：是啊。
莉亚·费格：加油啊，斯巴鲁。
佐伊·希弗：这就是我们今天的节目。我们会在节目说明中附上我们讨论过的所有故事的链接。《Uncanny Valley》由Kaleidoscope Content制作。阿德里亚娜·塔皮亚制作了本期节目。混音由Macro sound的阿马尔·拉尔完成。事实核查由丹尼尔·罗曼负责。普兰·班迪是我们的纽约录音室工程师。马克·莱达是我们的旧金山录音室工程师。金伯利·蔡是我们的高级数字制作经理。凯特·奥斯本是我们的执行制作人，凯蒂·德拉蒙德是《连线》杂志的全球编辑总监。
英文来源：
This week on Uncanny Valley, our hosts discuss Amazon’s controversial decision to drop Luca Guadagnino’s film about OpenAI’s Sam Altman—which reportedly did not paint him in a favorable light. Alongside Google DeepMind’s $75 million brand new partnership with indie film studio A24, how much of a dent is AI actually having in the films we see? They also dive into the recent upheaval of workers—from electricians to software engineers—against data centers. Plus: Meta’s program to track employees’ data gets paused after a massive leak, and Anthropic is now getting along with the government thanks to CEO Dario Amodei no longer being in the room.
Articles mentioned in this episode:
- A24 Knows You’re Mad About the Google AI Collab
- Some Electricians Think Building Data Centers Is for Sellouts
- Meta Pauses Employee-Tracking Program Following Internal Data Leak
- The Trump White House Is Over Anthropic CEO Dario Amodei
You can follow Brian Barrett on Bluesky at @brbarrett, Zoë Schiffer on Bluesky at @zoeschiffer, and Leah Feiger on Bluesky at @leahfeiger. Write to us at [email protected].
How to Listen
You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how:
If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for “uncanny valley.” We’re on Spotify too.
Transcript
Note: This is an automated transcript, which may contain errors.
Brian Barrett: Hey, this is Brian. Before we start, two quick things. If you've been enjoying listening to the show, we would appreciate it if you took a second to rate it in your podcast app of choice. It really helps us reach more people. And second, if you have any questions related to tech, privacy, or politics that you would like me, Zoë, and Leah to take on, now is the time to submit them to [email protected]. It doesn't matter how big or how small, we want to hear from you and get you answers. OK, on to the show.
Zoë Schiffer: Welcome to WIRED's Uncanny Valley. I'm Zoë Schiffer, director of business and industry.
Brian Barrett: I'm Brian Barrett, executive editor.
Leah Feiger: And I'm Leah Feiger, director of politics and science.
Zoë Schiffer: Today on the show: AI and Hollywood. We're discussing Amazon's MGM Studios’ sudden decision to drop the OpenAI biographical movie just as they were wrapping up production. At the same time, Google's DeepMind is investing $75 million to develop AI tools with the film studio A24. The AI and film industries are becoming increasingly intertwined and we're ready to take a look at where this is all headed.
Leah Feiger: We'll also dive into some recent controversies regarding data centers. As national and local backlash against data center construction increases, some electricians are even refusing to work on them and they're not alone. A group of Amazon workers claim they are being investigated for speaking out in favor of regulation.
Brian Barrett: And if you've been listening to the show in recent weeks or reading WIRED, you know that we've been covering the unfolding internal crisis at Meta very closely. This week we had another scoop, the controversial system that tracked employees every keystroke and screen activity has been paused after the company leaked sensitive data from it internally. We'll get into whether this series of frustrating incidents could actually lead to change within the company.
Leah Feiger: And later in the show, we'll get an update on how the talks between Anthropic and the government seem to be improving now that CEO Dario Amodei isn't in the room.
Zoë Schiffer: OK guys, I am so excited to talk about the business with this movie Artificial, which was suddenly dropped by Amazon's MGM Studios. So to get everyone up to speed, Artificial is a film by the director of Call Me by Your Name and Challengers, two great movies if I do say so. And it's a biographical drama about OpenAI and specifically The Blip, which was this moment in November 2023 when Sam Altman was abruptly fired by his board of directors and then swiftly rehired after basically the whole company revolted. I genuinely cannot believe they made an entire movie about this.
Brian Barrett: Yeah.
Zoë Schiffer: It's been described as The Social Network, but for the AI Age. The movie features a star-studded cast with Andrew Garfield as Sam Altman and Monica Barbaro as the former OpenAI CTO, Mira Murati. The movie was like mid-budget. I think they spent $40 million on production, but it was basically almost done when Amazon announced that they were dropping the film saying, "It would be better served if it were released by another studio." The decision has been drawing criticism because it's seen as Amazon basically doing a solid for Sam Altman, who the movie portrays pretty badly. I have to know what you guys think.
Leah Feiger: I have so many thoughts immediately. One, Andrew Garfield, we knew that. We knew it was always going to be him. Playing whiny tech bros is absolutely in his wheelhouse.
Zoë Schiffer: He famously played the Facebook cofounder who was kind of pushed out by Mark Zuckerberg in The Social Network.
Brian Barrett: I think too, there's some context here too, which is that Amazon has $50 billion invested in OpenAI, right? So in terms of like, when they say the movie will be better served by another studio, I think what they really mean is the studio will be better served by another movie, right? This is really like—
Leah Feiger: For sure.
Brian Barrett: —arm's length at this. What really strikes me about it is something that we've known is coming and has been happening, but the extent to which the film industry and the tech industry are intertwined right now. Amazon owns MGM. Paramount is being acquired by the Ellison family, Larry Ellison, obviously the founder of Oracle. So all these tech billionaires are now sort of totally intertwined with the movie industry, which is really going to determine what movies get made and what don't. And this is like a really glaring example of that.
Zoë Schiffer: Yeah. I mean, I can speak firsthand that there were many Twitter/Elon Musk film project that were in talks, but very difficult to get that to the screen for obvious reasons. I do think the film was, everything we heard was that it was going to be unflattering for Sam Altman. In fact, The Blip is pretty unflattering for Sam Altman because the reason that a lot of Sam Altman's executives turned on him and orchestrated what has been called a coup was that they perceived him to be duplicitous, to lie, to tell different people different things based on what he thought they wanted them to hear. And in the film, it really seems like Ilya Sutskever, who is the former chief scientist of OpenAI, really comes off as the hero. He's gone on to found another company that's really focused on safe artificial intelligence. And so it doesn't entirely surprise me that this was going to be a complicated project for Amazon to champion. In addition to the investment that you just talked about, Brian, they've also struck a $38 billion compute deal somewhat recently.
Brian Barrett: And Sam Altman was a guest at Jeff Bezos's wedding last year. It's personal, it's financial, it's all of these things.
Zoë Schiffer: Right.
Brian Barrett: And you sort of hope, like the idealized version of this, right, is that they would treat it the same way as they maybe treat the media industry when you have these billionaires taking over at these companies where it's sort of a hands-off, you do what you do. I trust the art. I'm doing this because I believe in it. That's not the case. It's less and less the case in the media industry as well. Jeff Bezos obviously has made some pretty big changes at The Washington Post, that there's sort of this exertion influence that you, in an ideal world, would not be happening. Powerful interests have owned Hollywood companies for a long time, but this is sort of just such an overt case.
Zoë Schiffer: Yeah. And I think that for OpenAI, they're really sensitive right now to public opinion. They feel like, and they're not the only ones, Anthropic feels this way too, that AI is increasingly unpopular. There's already been some talks about Sam Altman being ousted again before OpenAI goes public as soon as this year, probably next year. And so I think that while you might say, "Oh, they can just brush this off. They're focused on other things." They're used to bad media articles and perhaps negative portrayals. I think that they're increasingly trying to control the message.
Brian Barrett: And also the medium, right? I think the other deal that happened this week that drew a lot of eyebrows was Google DeepMind announcing a $75 million investment in A24, beloved indie darling, to create AI tools. Leah, I know you feel strongly and sadly about this.
Leah Feiger: Yeah. But the thing is I kind of hate A24.
Brian Barrett: Oh.
Leah Feiger: It's already so programmatic anyway. I'm like, ah, yes, another film in this exact sepia tone that has these exact actors playing these exact roles. The minute that I saw that it was A24 getting into bed with Google DeepMind, I was like, "Perfect. This was already an algorithm waiting to happen. Of course they're going to give it to the bots."
Zoë Schiffer: Well, this content is made for me. I'm a huge fan of A24. However, I've talked to a lot of people in Hollywood about this because there have been all of these deals, and yet we've seen OpenAI and Anthropic, like these big AI labs kind of hint that we will be able to produce films with AI really soon. And then you actually see an AI-generated film and you're like, certainly not. But what I am hearing is that AI is being used more and more for very specific purposes. It's things like storyboarding. And I believe that in the press release, Google DeepMind and A24 kind of mentioned that as a possible use case. It's things like rotoscoping, basically very labor-intensive, specific areas of the film development process that used to take a lot of human labor and were really expensive and you can genuinely automate. Those are, as far as some people think, really good use cases for AI. And I think we will start to see more and more shots, like specific shots on the big screen that are AI-generated. I think I would be very surprised if we see a feature-length film on the big screen that's AI-generated start to finish and is anything other than a one-off gimmick. There's an ick factor in realizing that something that you engaged with was AI-generated and you didn't know that I actually think is important not to dismiss. And then there's the like, can it actually create something that is quality enough to appear in that type of context?
Brian Barrett: I do think too, one thing that, this gets to your point earlier, Zoë, that when people see that there's a deal for $75 million, you assume that it's to train AI. You assume that it's going to be, oh, now Google's going to absorb all of A24 stuff into the borg and like—but it's not that. So it's interesting that it's still controversial obviously, but that it is really for those, like trying to find those purpose-built tools to support the creation of non-AI film, which gets lost and is, yeah.
Zoë Schiffer: You need a human. I think there was fear that Google would be training its models on the A24 catalog and that's specifically not part of the deal. I think probably Google wanted access to that IP, if I had to guess, but it would be a really bad brand move for A24 to give them that access.
Brian Barrett: We've talked about how everything's connected—file one more connection under this: A24 has Thrive Capital—is one of its investors. That is a venture capital firm owned by Josh Kushner, brother of Jared Kushner. Thrive has OpenAI as a major investment, holds a major stake in SpaceX, Warner Brothers about to be owned by the Ellison family—just a lot going on in terms of everybody owned by anybody related to everybody.
Leah Feiger: Film industry trying to cozy up to AI. There are other people that hate it for all sorts of different reasons too. Beyond creativity concerns, there is an ongoing pattern that we've been keeping an eye on, the sustained and increasing backlash against data centers across the country. Today, more than 40 percent of homes in the US are within five miles of an operating data center, according to a Pew Research Center analysis. And the construction boom just keeps going as the big AI companies devote billions and billions of dollars towards infrastructure to keep up with the demand. WIRED has reported about residents pushing back against data center construction in their communities from everything that it brings, from higher electricity bills to water scarcity, noise, you name it. So workers that are in theory, the ones implementing these projects are actually pushing back too in any way they can. Have you guys been following this?
Brian Barrett: Yeah, it's been interesting to see this evolution. I think we've talked a lot about communities pushing back and that remains an important story, but we've had a couple of stories on WIRED and there've been stories elsewhere about people who are directly involved. Caroline Haskins, a WIRED staff writer, had a great look at how electricians—right?—electricians are crucial to building data centers. Some of them are now saying, "Wait a minute, does this make you a sellout? Does working on a data center mean that you are kind of betraying broader principles, not just electrician principles, just human principles?"
Zoë Schiffer: You're missing the best part of that story, which was when the electrician said that it's really hard to date once he tells people that he's an electrician who works on data centers.
Brian Barrett: It's like Cybertrucks. It's like owning a Cybertruck.
Zoë Schiffer: It really did remind me of that.
Brian Barrett: Working on a data center is the new owning a Cybertruck.
Zoë Schiffer: The difficult thing about this, I think, I mean, so yeah. People don't want to live near data centers. They spike your energy bills, they're loud, and bright, and take all sorts of resources. I think that makes a lot of sense. In terms of countrywide economic initiatives, data centers in some ways feel like all we have. Am I going to get skewered for saying that? I'm just like, this is like the big thing, AI is the big thing that America is betting on and I think it's complicated to try and slow that down or roll that back rather than try and push for them to be less horrible to live near.
Brian Barrett: But I think here's the thing. I agree. The building of data centers and investment in AI is the US economy at this point. And when that goes away—
Zoë Schiffer: Right.
Brian Barrett: It's going to be really, really painful. I think a lot of the pushback though, which is that there's not really much direct economic benefit for you locally if a data center comes. Usually when you have like, oh, they're bringing a railroad into town, at least you get added commerce from the railroad, right?
Zoë Schiffer: Right.
Brian Barrett: This is what happened.
Zoë Schiffer: Oh, my God.
Brian Barrett: I'm old enough.
Zoë Schiffer: This is the one railroad example that I'll accept.
Brian Barrett: Thank you. Yes, good. I've made it. So I think that's part of it. I think at the same time, you've got people inside these companies who are leading the charge. So it's not just people directly affected. It's people who actually stand to benefit from this stuff, both the electricians, people inside Amazon. For example, some Amazon employees recently urged the Seattle City Council to regulate data centers. So even people with presumably a financial interest in this stuff are pushing back. I get what you're saying, Zoë, but I think there is a certain—it makes sense to me that people are saying, "Wait, hold up," because it's not an economy that everybody's able to participate in.
Zoë Schiffer: No, totally. I want to be clear. I completely understand the pushback. If one of these were built in my neighborhood, I would be pushing back. I think it's totally legitimate not to want to live near them. I just think the situation is complicated when so much of our economy now hinges on this industry. It's not a reason not to regulate, but I think it's going to be very complicated to roll it back. I've also been curious, and Leah, I'm curious on your perspective on this, just to see which politicians have really—I was kind of struck that Bernie Sanders has become a really vocal opponent to data centers.
Leah Feiger: Yeah. Senator Bernie Sanders and Rep Alexandria Ocasio-Cortez, AOC, they introduced the Data Center Moratorium Act, which would halt the construction of new AI data centers until there were actual national safeguards. This sounds, in so many ways, like a very baked for left wing issue, but shockingly, this is pretty bipartisan. There are folks from all sides of the aisle getting involved because their constituents are reaching out and going, "What are you doing in my name? What are you doing in my backyard? How is this benefiting me? How is this hurting me?" So I think the thing that I've been most shocked by is the bipartisanship of this.
Zoë Schiffer: It's interesting because I feel like, and this is purely my speculation, but just based on how OpenAI talked about data centers, really came out in front during the first day of the Trump administration, kind of championing big data center build out projects. I was like, I'm reading Chris Lehane, the company's chief global affairs officer, previously very high up at Airbnb and a political fixer before that as like this was something that he and OpenAI might've thought was going to be really beneficial for the company. It was like an America first build, baby, build kind of message.
Leah Feiger: We're giving jobs to everyone.
Zoë Schiffer: They just misread the moment. They did not realize how toxic this issue was going to be. And now it's very hard to kind of change their stance when they've been releasing press releases every time there was a new data center. Now it's like, "Uh-oh, we got to keep it quiet because people really don't like this."
Brian Barrett: And to your point, they can't roll it back either way until you can put data centers in space because they need the compute.
Zoë Schiffer: Which by the way, is going to be really difficult to do.
Brian Barrett: If not, impossible. Zoë, is there any chance that the sort of internal dissent, right, electricians saying, "I don't think so." Workers inside of companies saying, "Hey, we don't like data centers either." Any chance that that changes anything at all in terms of the trajectory for these buildouts, for these companies, for the spending?
Zoë Schiffer: I would be very surprised. I don't want to say absolutely not because we have seen examples where famously Google workers all came together, pushed back on Project Maven, some of the censored search projects for China and what have you, and actually got those launches paused.
Brian Barrett: Just real quick, Project Maven was working with the Pentagon basically, right, using Google Tech for the DoD?
Zoë Schiffer: Exactly, exactly. So yeah, it's happened before. It could happen again. What I would say is two things. One, the pushback we've seen from the hourly workers has been minimal when you look at the entire workforce. They are bringing in thousands of peoples. I've heard that they're paying much higher rates than people typically get on these jobs. And so I think for an industry that has historically needed a lot of work, I think there will be people who are willing to work on these projects and then we'll hear little pockets of dissent and pushback, which again is newsworthy, and relevant, and it's not no one, but I still think they're able to hire thousands and thousands of people. I would also say that on the corporate level, while we are starting to see more pushback, more vocal opposition from corporate workers in terms of what their companies are doing, it's still at a far lower level than it was around 2018.
Brian Barrett: Especially when the job market is radically different now for these engineers. And I think a few years ago there was a turning point too where Google fired a bunch of people for protesting, whereas before they had sort of allowed it, task at least sort of let it happen. I will say there is one thing that will get management to listen to employees and that is a colossal screw-up, which we saw at Meta recently. We've talked before about how Meta installed software on employee devices to track every keystroke, all the screen activity, basically full surveillance on its employees to help train AI. This week, WIRED broke the news that the company apparently left potentially sensitive information from those sessions exposed and accessible to anyone inside Meta. So if you're a Meta employee, everything you did on your screen could be viewed by any of your coworkers. Can you imagine?
Zoë Schiffer: No, and I feel like we have to just back up a little further even and lay this out for people. So Mark Zuckerberg feels like he is behind in the AI race. He invests billions of dollars in building up this new AI lab, spending so much money on specific talent, giving them so many resources to catch up and build frontier models. Then the company says, "We're spending so much on the AI buildout that we actually have to lay off a whole bunch of you, 10 percent of the workforce." Then they say—
Brian Barrett: 8,000 people.
Zoë Schiffer: Right, 8,000 people. Then they say, "By the way, you're going to have surveillance tech installed in your laptop to track your keystrokes, whether you like it or not, to train the models that that fancy AI lab that just was connected to all these layoffs are trying to build. And then, oh, by the way, we left all of that data exposed accidentally."
Brian Barrett: But we will be holding a hackathon in a few months to increase morale. The one thing that came out of this that is maybe a silver lining or a good thing is that Meta did announce that they are pausing the data gathering program while they investigate what exactly happened here, but it's a pause. It's a pause.
Zoë Schiffer: They did say pause. Then we updated that story later that very night and they made it very—they had not said anything company-wide about the leak. We had been hearing about it from a bunch of different sources. They were still keeping it pretty locked down. Then late at night, we finally got the message that they had sent out and they very much said, "We're going to evaluate what happened here, make sure it doesn't happen again before we relaunched the tools," was the message.
Brian Barrett: The guy who's been the face of a lot of those apologies is Meta CTO, Andrew Bosworth, known as Boz and it sort of burns a little bit extra because before this happened, before this internal leak happened, people had asked him straight up, "Isn't this potentially dangerous from a privacy perspective?" And he told employees, "Actually, it's tightly controlled and we use the same protection standards, and store systems, and access controls as other sensitive data sets," which makes you worry about the other sensitive data sets—
Zoë Schiffer: Truly.
Brian Barrett: That Meta is sitting on.
Zoë Schiffer: Completely. I'm excited for the feature that we assign that's titled Boz and his Metamates. I feel like Brian, I don't know, it just seems like a story that you would say yes to.
Brian Barrett: It's right. Can't stress enough that Metamates is what people call each other inside of Meta. Do you guys want to hear a leaked audio of Mark Zuckerberg discussing the training initiative with employees last month and why he thought it was essential?
Leah Feiger: Always.
Zoë Schiffer: Absolutely.
Mark Zuckerberg, archival audio: We're in a phase where basically the AI models learn from watching really smart people do things. In general, the average intelligence of the people who are at this company is significantly higher than the average set of people that you can get to do tasks. So if we're trying to teach the models coding, for example, then having people internally build tools or solve tasks that help teach the model how to code, we think is going to dramatically increase our model's coding ability faster than what others in the industry have the capability to do who don't have thousands and thousands of extremely strong engineers at their company.
Zoë Schiffer: You're actually so good at taking meeting notes, that you're just, you're great at it. So we had to give you that job.
Brian Barrett: You all are going to be so good at replacing yourselves with AI, that we're going to let you do that even faster.
Leah Feiger: Can we talk about another AI company, you guys?
Zoë Schiffer: Oh my gosh, Leah.
Leah Feiger: Your dream, Zoë, you're dreaming—
Zoë Schiffer: You go away—
Leah Feiger: Of me saying that.
Zoë Schiffer: For a little bit of time and you're ready to just talk AI all of the time.
Brian Barrett: One of us. One of us.
Leah Feiger: Well, it has to do with Anthropic and the US government. As we know, Anthropic has had a bit of a dust up with the Trump administration over their most advanced models, Fable 5 and Mythos 5 after the National Security Agency affirmed that there were ways to jailbreak and disable the security guardrails in these models. Since then, both sides have been trying to find a way forward. Sometimes there has appeared to be movement, other times less so. But recently, WIRED's Hugo Lowell reported for his newsletter Inner Loop that the administration has had multiple calls with Anthropic in recent days, appears to be making some moves, some encouraging moves, but it's mainly all due to the fact that they don't have to interact with the company's CEO anymore, Dario Amodei. This quote is going to stay with me forever, you guys, because the administration is very excited that Tom Brown, Anthropic's cofounder and Anthropic's public policy chief, Sarah Heck, have been leading the outreach. This is the quote about this that we have from an official. I'm never going to get over it. "Tom Brown is not being a weirdo like Dario and can actually engage."
Zoë Schiffer: Incredible. I mean, no one was ever saying that Dario wasn't very weird. I mean, that man is an oddball.
Brian Barrett: I love this quote for us. I feel so bad for him.
Leah Feiger: I feel so bad for him, but it was also just this incredible like, what's changed here. No, no, it's not like the technical issues. It's the who the interactions are with—
Zoë Schiffer: Right.
Leah Feiger: Which is, I don't know, perhaps the Trump administration in a nutshell. These talks have been happening at the high level, working group level, technical staff.
Zoë Schiffer: I feel like we need to lay out those stakes really quick for people who haven't been following every twist and turn of this saga, which wow, your brain must be so healthy and I'm jealous, but.
Brian Barrett: It must be nice. Yeah.
Zoë Schiffer: That must be nice. The US government, after the NSA basically alleged that there were ways to jailbreak the models, implemented export controls on those models and basically said that it was not OK for Anthropic to allow foreign nationals to access the model. There is no technical way for Anthropic to do that. And so what they had to do was roll back access to both of those models for everyone. It's a really big deal for the company. And so these talks have basically been trying to hash out from Anthropic side, what do we need to do to get those models back on the market? And from the US government side, what do we need from Anthropic to feel comfortable that these models can launch and they won't be a huge cybersecurity nightmare?
Brian Barrett: Well, especially it seems as though also like the Trump administration from our understanding is that they're asking for something that's basically not possible. They're asking for a model that you cannot jailbreak, which is just not a thing. The nature of these models, the best you can do, which Anthropic has said, "Hey, we can do this," is to sort of make it nearly impossible to do a universal jailbreak where you can get it to do anything you want it to do. But you can see how for a conversation, especially like this, having someone who can talk about these issues on a non-technical level and on a political level and someone who's maybe able to suppress their highfalutin ideas about AI, you can see how it really matters in situations like this and is going to continue to matter for all of these companies. They're all going to have to go in front of the White House at some point and make a case for why this super powerful model should be released.
Zoë Schiffer: You really need someone who can get through an explanation without saying Fermi's paradox, orthogonal. They got to be able to speak the English language.
Leah Feiger: So is that what this is going to come down to? Are national security concerns alleviated by someone that actually listened to their PR person for longer than three minutes about how to talk in public?
Brian Barrett: I mean, yes. I'll go even further. If they had listened to their PR person two months ago or if their PR person had taken a different tack, I think we wouldn't be here in the first place.
Zoë Schiffer: Yeah. I mean, but this is the interesting thing because Anthropic has spent so much time publicly saying, "AI is really dangerous. AI could end humanity. Our models are so scary, and so good at cybersecurity, and hacking, that we can't release them publicly." And so I do feel like they have, and Dario Amodei in particular has positioned himself and the company in this very specific position, actually also saying that we need a lot of government regulation. When they're saying all those things and they don't have strong relationships with the Pentagon, they are walking a very, very fine line. It is not that long ago that there was this big falling out because Anthropic was trying to put a line in the sand and say, "You can't use our models for any type of military action that you want. You can't create autonomous weapons with them," or what have you. And the Pentagon was saying, "You don't get to decide what we do with these models. This is American AI and we want to do what we want to do."
Brian Barrett: Coming up after the break, we'll share our WIRED/TIRED picks for the week. Stay with us.
Zoë Schiffer: It's time for our WIRED/TIRED segment. You know the drill, whatever is new and cool is WIRED, whatever passé thing we're over is TIRED.
Leah Feiger: OK. I'm going to start on a positive. My WIRED is Caroline Calloway. She is—
Brian Barrett: Wow.
Leah Feiger: Back on our feeds, back on our feeds people, just going through a terrible breakup and posting through it.
Brian Barrett: Give people two sentences on Caroline Calloway because I have been online too much for the last decade, but not everyone has been. So give them a little bit of who this is and why it's so exciting.
Leah Feiger: Caroline Calloway is like an erstwhile grifter now downtown Dime Square gal. I encourage you, if you don't know who this woman is, to jump down the rabbit hole. Very, very fun. Couldn't recommend more.
Zoë Schiffer: So she's WIRED.
Leah Feiger: Yeah, she's my WIRED and my TIRED is Feed Me's Emily Sundberg.
Brian Barrett: Leah's just going for it.
Zoë Schiffer: We're torching.
Brian Barrett: Going for it.
Leah Feiger: We're torching. Zoë has known I've had inklings of this for a while. Emily Sundberg's Substack used to be one of my favorite Substacks. I think that she was doing something very, very cool. She was making business news, and tech news, and cultural news. Recently, in the last couple weeks, months, a lot of people have also really taken note of Emily Sundberg. She has gotten a lot of press and collaborations and I'm trying really hard to phrase this correctly and kindly anyway, but her recent newsletter from this week where she interviews Mark Zuckerberg from Meta about his style and time at Harvard made me want to die. So TIRED, Emily Sundberg. I'm so sorry. I think you could pull it back out. I really do, but everything you write is embarrassing now and I'm over it.
Zoë Schiffer: OK. That's a Leah-specific opinion, Emily. I still love your newsletter.
Brian Barrett: Leah does not necessarily speak for Uncanny Valley as a whole.
Zoë Schiffer: No, we're very divided actually. We've realized during the course of this pod. All right, Brian, let's go to—
Brian Barrett: No, I can't follow that.
Zoë Schiffer: I hope you have the most tepid WIRED/TIRED ever.
Brian Barrett: I can't follow that. No. WIRED is making friends and TIRED is keeping—no, mine is boring compared to that, but I'll say it anyway. Leah, you should have gone second. You should have insisted. My WIRED is cheap EV trucks. Cheap EV trucks. There I said it. Don't look at me like that. The Slate is a new EV truck that's been in the works for a while. It's had a lot of hype behind it, but they're going to sell it for $25,000, which is still a good chunk of change. But the average new car price in the US, does anyone know what it is? $50,000 is the average new car price. So they're coming in with an electric truck. It's all gray. It's very boring. It is no frills, but it's $25,000. So it is an accessible electric truck that anyone can buy. And the reason they're able to do that is because this car has nothing. You have to crank the windows, you have to pay extra for every—it's like the Spirit Airlines of EVs where you have to pay for—
Zoë Schiffer: Wait, this is TIRED. We don't want this.
Brian Barrett: No, this is WIRED. I love this. I think it is WIRED, that there is an affordable EV that you can buy and then you can just add, there are over 200 accessories that you can add to this. So if you want other things on this car, you want the car to have a color, for example, or you want automatic windows or whatever, you can add that piece by piece and distinct to each car instead of automatically you walk in and have to pay $50,000 for stuff you may not want or need. I think these days, especially as the government is pulling back incentives for EVs, there's this big push toward oil. To see something that people can actually buy and actually customize, that actually has decent range, I think is really nice, unexpected. That's my WIRED. TIRED is whatever the opposite of that is.
Zoë Schiffer: I know when you said Spirit Airlines, I was like, that sounds like a negative. I do really want to get the Rivian as my next car. I'm all in on the electric. Yeah.
Brian Barrett: I'm waiting for the R3, which apparently won't come out until 2030.
Zoë Schiffer: Oh my gosh. Yes.
Brian Barrett: Hang in there.
Leah Feiger: I just want Subaru. I want a Subaru EV. I'm putting that out into the universe.
Brian Barrett: Yeah.
Leah Feiger: Come through Subaru.
Zoë Schiffer: That's our show for today. We'll link to all the stories we spoke about in the show notes. Uncanny Valley is produced by Kaleidoscope Content. Adriana Tapia produced this episode. It was mixed by Amar Lal at Macro sound. It was fact checked by Daniel Roman. Pran Bandi is our New York studio engineer. Marc Leyda is our San Francisco studio engineer. Kimberly Chua is our senior digital production manager. Kate Osborn is our executive producer and Katie Drummond is WIRED's global editorial director.
文章标题：亚马逊放弃其OpenAI电影项目的原因、数据中心员工发起抵抗，以及Meta泄露员工数据
文章链接：https://news.qimuai.cn/?post=4438
本站文章均为原创，未经授权请勿用于任何商业用途