今天在《解码器》节目中,我与 The Verge 资深 AI 记者海登·菲尔德聊了聊苹果与 OpenAI 之间那起重大的商业秘密诉讼,以及这起案件揭示了 OpenAI 怎样的未来。
我相信现在大多数《解码器》听众都已熟悉苹果在这起案件中的指控。该公司称,OpenAI 的数名前苹果员工在求职面试中瞄准了苹果的商业秘密,甚至从苹果服务器下载了与硬件制造相关的文件。这些指控令人震惊,尽管 OpenAI 予以否认,且我们尚未看到该公司的法律回应。
海登多年来一直在报道 OpenAI,她刚刚还与多位专攻知识产权和商业秘密法的律师进行了交流,以了解这起案件的严重程度。答案对 OpenAI 来说并不完全乐观——该公司曾在 2025 年斥资 65 亿美元收购了传奇苹果设计师乔尼·艾维的 AI 硬件初创公司 io Products。
正如你将听到海登解释的那样,苹果是出了名的难缠的诉讼方,而且我们有大量科技行业历史可以佐证这一名声。但那些案件针对的是微软和三星这样规模的公司,是能够负担得起此类诉讼成本和干扰的巨头企业。
OpenAI 还不是那样的公司。它仍在烧钱,在战略上有些迷失方向,同时试图在企业市场和消费者市场展开竞争,并且高管层频繁更替。因此,海登和我花了大量时间讨论这起案件可能对一家定义了 AI 热潮早期阶段、但近来似乎总是陷入各种危机的公司意味着什么。
好的:The Verge 资深 AI 记者海登·菲尔德将带来关于苹果与 OpenAI 诉讼案的报道,以及 AI 行业可能因此发生怎样的变化。我们开始吧。
海登·菲尔德,你是 The Verge 的资深 AI 记者。欢迎回到《解码器》。
谢谢。很高兴来到这里。
又一次,我们要聊聊 OpenAI 了,海登。
又一次,它永远停不下来。
我们来聊聊 OpenAI 能否在极度分心的时刻保持专注——这似乎成了这家公司的一个老话题。不过这一次,分心的源头并非来自他们内部,而是来自外部世界,或许正是他们自身某些行为所引发的。
当然,我们说的就是那起苹果商业机密诉讼案。这感觉是件大事。苹果指控 OpenAI 通过各种方式窃取该公司信息,甚至在面试环节——要求面试者透露他们对苹果产品的了解。请简要介绍一下这起诉讼以及目前的情况。
当然。这确实很疯狂。我上周写了一篇文章,标题是《萨姆·奥尔特曼不需要再多一场官司了》,而这正是目前的真实处境。从苹果提出的一些指控来看,情况似乎相当恶劣。据称,一名前苹果员工在面试中涉嫌套取商业机密,要求应聘者将苹果硬件带出办公室,带到 OpenAI 进行“展示与讲解”之类的操作。
指控内容很多,相当离谱。但当我上周实际采访多位专家时,他们表示这些情况他们以前见过很多次了。只不过,他们从未见过所有指控集中在一个案件里,而且涉及两家如此关键的公司。
如果这些指控属实,其中一些内容确实相当恶劣。但与此同时,该领域的专家们也表示:“哦,我们在商业机密案件里经常见到这种情况。这并不疯狂。真正疯狂的是,所有指控都集中在一个案件里,而且涉及两家规模如此庞大的公司。”
这确实很有意思。这或许是一种模式,虽然不值得深入探讨,但在 90 年代,苹果曾将版权法推到极限。他们起诉微软,指控 Windows 抄袭了 Mac 的外观和感觉。我要指出的是,Windows 至今依然存在,而微软也安然无恙。
到了 2000 年代和 2010 年代初,苹果将专利法推向了极限,起诉三星涉嫌因 Android 和 Galaxy 手机侵犯多项专利,因为苹果认为他们保护了 iPhone。三星最终支付了约 10 亿美元的赔偿金,而这些案件也一直无休止地循环往复。
但我要指出的是,Android 依然存在,三星也发展得很好。苹果或许重新定义了围绕 3G 和 4G 许可的专利法,但从根本上说,它实际上并未成功通过知识产权诉讼阻止其竞争对手。
所以,从版权到专利,再到现在的商业秘密——这是第三大知识产权领域——苹果又在 OpenAI 身上故技重施。我最大的疑问是,OpenAI 能否像微软和三星那样直接付钱了事。看看三星,支付 10 亿美元的罚金来让 Galaxy 手机起步并成为移动领域的主要竞争对手,这可能是该公司有史以来最轻松、最划算的投资。
但我不确定 OpenAI 是否处于与当年三星相同的阶段。从苹果目前告诉我们的情况来看,似乎所有跳槽到 OpenAI 的苹果员工可能都越界太多了。
读到诉讼中最夸张的指控以及那些交流信息,实在令人震惊。其中一条指控称,有人在一台设备上访问了苹果的机密信息,然后说:“哈哈!我能访问这些信息,太搞笑了。”这里面信息量很大。
我也一直在想同样的问题。这真的取决于苹果是愿意斤斤计较,还是选择极其理性地处理。当我就此采访多位律师时,他们表示,苹果是出了名的难缠诉讼方,正如我们在过去那些案件中看到的那样。而且,他们确实会把事情闹大。他们资金雄厚,也愿意为此花钱。这真的只取决于他们有多生气。
我认为,如果 OpenAI 付给他们足够的钱,他们可能会和解,但 OpenAI 能做到吗?他们今年已经非常艰难了。他们即将上市,所有投资者都在盯着他们的利润表,想知道他们何时才能盈利。现在要拿出一大笔钱给苹果,时机很不好。我们得看看他们怎么做。
我认为这件事会进入庭审,并演变成一场旷日持久、持续数年的拉锯战。我采访过的许多律师也是这么说的。但我也可以想象这样一种情况:他们想办法凑齐这笔钱,然后苹果可能就接受了。
在三星的专利案中,官司打了很久。在那期间,蒂姆·库克出席了一次 Code 大会,我问他:“你在做什么?这有什么意义吗?”他非常严肃地看着我说:“嗯,我们认为标准必要专利的许可费太离谱了。”我当时心想:“这根本不是你做这些事的理由。重点不在于此,重点是你很生气,你想把这个官司打到底,而且无论名义上的理由是什么,你都会把它打到底。”
这感觉像是整个拼图中的一大块。OpenAI 挖走了不少苹果员工。众所周知,苹果最著名的设计师乔纳森·艾维正在与 OpenAI 合作开发硬件设备。还有唐坦,他是 OpenAI 的首席硬件官。他在苹果公司时就有“行事不严谨”的名声,这是彭博社马克·古尔曼的说法。
他似乎是这个故事的核心人物,与其说是乔纳森·艾维,不如说是唐坦。而那一部分——一群前苹果员工可能行事过于随意——几乎让人觉得这是个人恩怨,几乎感觉其中掺杂了很多情绪。谈谈唐坦,谈谈所有这些被指控的行为,感觉这比一般的大型企业纠纷更带个人色彩。
关于你提到的与蒂姆·库克的对话,我真心认为人们低估了企业的狭隘心态。近年来我们已经看到不少例子,比如 OpenAI 与埃隆·马斯克之间的诉讼。如果企业不是一味地想要较劲,很多事情本可以避免——当然,较劲也没什么不行。他们有这个资本,钱多到可以这么做。但人们往往低估了这种心态在这些事件中有时所起的作用。
至于唐·坦,是的,他在苹果公司工作了 24 年,曾是 Apple Watch 的副总裁。他于 2024 年离职,加入了乔尼·艾维的硬件公司 io Products,而这家公司后来显然被 OpenAI 收购了。
随后,OpenAI 任命他为首席硬件官。因此他在 OpenAI 担任了高管职务,但这发生在他为苹果效力 24 年之后。根据苹果诉讼中的指控,他是大量商业秘密收集活动的幕后策划者,涉及他主持的面试——在面试中他会询问一些秘密代号项目的情况。而接受面试的人可能会觉得,既然他能知道这个项目的名称或相关细节,那说明他本人也参与其中?于是,据称他们便认为自己可以更自由地畅所欲言。
还有其他时候,他会问:“嘿,你为什么不把以前在苹果办公室外做过的电池或其他硬件零件带一些过来,我们好做个展示讲解?” 梳理这些指控非常有意思。如果这些指控属实,那看起来他当时似乎真的毫不担心这会带来什么后果。
这里的另一个关键人物是乔尼·艾维,苹果那位与史蒂夫·乔布斯合作过众多著名项目的知名前设计师。据称,艾维与苹果的关系非常紧张。
如今他已经离开,并在各种采访和声明中明确表达了这一点。他正与山姆·奥尔特曼合作开发新设备。他还与山姆·奥尔特曼一起拍了一段视频,我只能将其形容为一场浪漫的咖啡约会。
完全同意。
在视频中,他们将手机和笔记本电脑描述为过时的设备。
这段视频立刻成了网络迷因。不得不说,那些迷因真的太精彩了。
这东西非常好。而他们为了打造它,封锁了旧金山半个城。我觉得艾维在那段视频里把手机和笔记本电脑称为“过时设备”,简直就是在直接针对苹果。这些产品是他亲手创造的,他在很多很多方面彻底革新了它们。而他却说:“这些已经过时了。”
关于这种说法是否属实、能否成真,你可以说很多,我也想稍后聊聊这个话题。但让我注意的是,这起诉讼中根本没有提到乔纳森·艾维。苹果有一条底线不会逾越,那就是提及这样一个事实:是乔纳森·艾维去了 OpenAI,开始招募他所有的老设计师,并将自己公司的大部分股份卖给了 OpenAI。而现在,他们正试图制造硬件,并涉嫌窃取商业机密。但乔纳森·艾维,他并不在其中。我们干脆不提他的名字。
是因为他太出名了吗?是因为他太敏感了吗?是因为他和劳伦·鲍威尔·乔布斯还是朋友吗?你怎么看?
我认为更可能的是,他知道如何抹去自己的痕迹。也可能是因为他势力太大,或者你提到的其他原因,但我真的觉得,他只是让别人替他干那些脏活。这起诉讼中的其他内容相当恶劣,而且他们全都查到了。所以,如果他们真的掌握了乔纳森·艾维的一些严重证据,难道会不写进诉讼里吗?他们之间是有过节的。我觉得他们肯定会写进去。
这只是猜测,但当你身处足够有权势的位置时,你不需要亲自做所有这些事。你有别人替你去做,而且你也知道如何抹去痕迹。所以,在我看来,也许他们只是没有掌握他的任何把柄。
是啊,或者他正忙着设计没人喜欢的法拉利,或者随便他在做什么,反正他没掺和进来。我确实觉得这很值得注意,因为山姆·奥尔特曼聘请乔纳森·艾维的目的,就是为了让下一款设备拥有可信的说服力。无论那款 AI 驱动的设备属于什么品类,你都要请来最著名的设计师,让他说:“我造出了 iPhone,而我将要造出这个东西。”
而且看起来 OpenAI 似乎也想要苹果的大量技术诀窍,才能做出那个东西。你怎么看这件事——OpenAI 需要苹果的硬件技术诀窍?或者至少苹果声称,OpenAI 需要苹果的硬件技术诀窍才能做出下一个产品?
在某种程度上,无论你谈论的是商业机密,还是通过雇佣其员工获得的合法公开技术诀窍,他们当然需要这些。他们看到大量 AI 硬件项目失败得一塌糊涂,正努力避免成为下一个 Humane AI Pin。在当今时代,做一款 AI 硬件设备可能会极其尴尬。你很可能惨败,沦为笑柄。所以,如果你想避免这种情况,当然需要把最优秀的人才拉到自己这边。
硬件很难做,这是公认的,所以他们当然希望引入那些在硬件领域真正取得过巨大成功的专家。这对我来说并不意外,而且我认为这是一个明智的决定。OpenAI 过去已经证明——这也是所有科技公司的通病,并非 OpenAI 独有——他们有时会与消费者的实际需求有些脱节。
你推出一个产品,没人真正想要。好吧。对于软件来说,调整方向相对容易一些,就像我们看到 OpenAI 砍掉那些“支线任务”一样。他们会说:“哦,这个我们不做了。那个也不做了。这个我们合并到那个里面去。”但硬件不行。你必须从一开始就做对。
很高兴你提到了消费者,因为这引出了我认为对 OpenAI 而言的核心问题:它本来是要成为一家面向消费者的 AI 公司。Anthropic 以专注于企业市场而闻名,并且这对他们来说利润丰厚;也许钱都在那里。但 OpenAI 原本是要通过 Bing 或者自己的应用来击败 Google 的。它本来是要成为一家消费者公司的。
你之所以要做硬件,唯一的目的就是去进攻消费市场,去进攻 iPhone。所以,OpenAI 的野心,至少在我看来,首先是抢占谷歌的市场份额——这是人类历史上最赚钱的生意之一——然后去进攻 iPhone,这是有史以来最著名、文化上最具统治力的科技产品。
我不知道它怎么能做到其中任何一件事。消费者似乎并不想要这些产品,也不想要这种规模的 AI。我不知道消费端的收入从何而来,而且大学生们都在毕业典礼上对 AI 喝倒彩。OpenAI 是不是只需要转向企业市场?它是不是需要在搞清楚自己的商业模式之前,先在消费硬件设备上白费力气?
OpenAI 内部已经多次转向。它最近的一次转向是,它将专注于其关键收入驱动力,即企业和编程。他们已经说过一百万次了。OpenAI 放弃了 Sora、Atlas、以及色情聊天功能。感觉像是几年前的事,但疯狂的是,这些都发生在过去一年里。
他们已经做了很多这样的事情,所以每个人都认为,“好吧,他们要全力以赴了,他们要搞企业市场了”,因为这是经过验证的赚钱方式,尤其是在他们即将 IPO 的一年里。公平地说,是的,他们早在这次全面转向和砍掉支线任务之前,就已经计划好要做这个硬件设备了。我猜他们承担不起砍掉这个潜在的支线任务的代价,因为他们已经在这上面花了很多钱,无论如何都要做下去。所以现在他们只是在尽力让它成功。
我报道 AI 领域才六年。在 AI 行业里这算很长了,也许我漏掉了什么,但我就是看不到一个每个人都在购买 OpenAI 硬件设备的市场。如果我错了,我们可以回到这里,嘲笑我 2026 年 7 月的这段视频。一个智能音箱,好吧,也许人们买它只是为了当个像 Google Home 那样的智能音箱,我自己就是这么用的。
我认为这会是那种少数人极其、极其、极其喜爱,但普通大众很可能不会接受的事物——因为除非他们所有的朋友都在用,否则很难让他们接受任何新东西。这里面有社交因素。如果你没有 iPhone,发绿色短信就会成为笑柄。没有冒犯安卓用户的意思,我只是说有时候你可能会被排除在群聊之外。这很可悲,也不公平,但很多时候,要推广一款新硬件,必须获得公众的认可。
我们在智能手机热潮之初创办了 The Verge。回想起来,人们是否会接受智能手机从来就不是一个问题。那是显而易见的。对我们来说显而易见,也许对所有人来说并非如此,但我们身处其中,唯一不明确的是谁会胜出。
过了一段时间,苹果会胜出也变得显而易见了。15 年前我们报道过多少款“iPhone 杀手”?它们来了又走,看起来都像新的操作系统。Windows Phone 存在过。Palm Pre 运行一个叫 webOS 的操作系统。安卓正在崛起。诺基亚试图走自己的路,然后转向了 Windows Phone。
最终,我们只剩下 iOS 和安卓,其他一切都靠边站了。也许整个科技行业都讨厌这一点,讨厌支付 App Store 的费用,但感觉现在又到了类似的时刻:OpenAI 将推出一款硬件设备,运行自己的操作系统。埃隆·马斯克正在大肆宣扬 Starlink 和 xAI 要推出自己的手机,可能也运行自己的操作系统,据说在 IPO 前夕,他们曾向人们展示过。
我只有一些非常愚蠢的问题,在我看来,这些问题与 15 年前我们关于新智能手机操作系统提出的问题如出一辙。你会做一个 Gmail 客户端吗?它会有 Instagram 吗?因为如果没有这些东西,你实际上无法取代手机。而如果你取代不了手机,那么,我手里的 iPhone 上已经装了 iOS 27 的测试版,Siri AI 绝对足够应付日常使用了。
这正是让我在意的地方,因为智能手机取代了功能机。所以,好吧,你得拿出比智能手机好得多的东西。你能做到吗?我表示怀疑,但咱们走着瞧。
但另一件让我在意的事是,据报道 OpenAI 正在研发的设备数量之多。据说有五款硬件设备,而无屏智能音箱是第一款。他们怎么能想出五款来?这正是我想知道的。这五款分别是什么?难道不应该把所有资源集中到一款上吗?做出一款就已经很难了。我对五款成功的 AI 硬件设备有点怀疑。哪怕只成功一款都希望渺茫,但五款就有点疯狂了。
我确信会有很多产品,有些会来,有些会走。我想到的是,尤其是乔纳森·艾维及其硬件设计方法——这一点大家已经很熟悉了,我们看过他在苹果发布过很多硬件——艾维倾向于为新设备寻找新的输入方式。
iPod 有触控滚轮,这真正定义了 iPod 的形态和功能。iPhone 有多点触控屏,这真正定义了 iPhone 以及后来 iPad 的形态和功能。如果你还记得 Apple Watch,他们发布时大张旗鼓地宣传了侧面的数码表冠。我不知道这个设计后来是否成功,但他们在发布 Apple Watch 时确实认为那是个大卖点。
这就是艾维的惯用路径:“我们找到一种新的输入机制,然后围绕它来定义产品。”我很清楚这对 AI 来说应该是什么——那就是自然语言输入。你只需告诉电脑你想要什么,它就能理解并为你执行。
但我们已经进入 AI 革命多少年了?这个版本有三四年了吧?它还没完全奏效。我认为这就是 AI 设备不断失败的原因,正如你一直提到的。因为你可以对着 Humane Pin 没日没夜地说话,它会假装理解你,但实际上却无法完成背后的那些任务。
这感觉就是当前 AI 领域的问题所在。一方面有灵光乍现的亮点,另一方面又存在能力上的局限。你觉得我们目前正处于弄清楚它究竟能做什么、不能做什么的哪个阶段?而这种局限性又会对硬件设备产生怎样的制约?
我们仍处于探索过程中。我个人一直在使用 AI,当然是在个人生活中用来测试各种功能,看看它能做什么。逐月来看,我至少从自己的个人体验中看到了它的进步——当然这些都只是零散的个人感受。
但就在前几天,我想查很多公司的地址,因为要绘制一个地图。我查到的地址大概有 60% 是完全错误的,就是错的。它给我生成了一整张表格,还说:“放心,这些肯定都是对的。”结果 60% 是错的。我说:“这里面很多都是错的。”它就说:“哦,抱歉,是我的错。”
对,我最喜欢这种反应了。
就像在说:“哦对,你说得对。被你抓到了。”
它在很多时候、很多事情上确实是对的。但正如我们当年当实习生时,或者带过实习生的人都知道,如果你不能百分之百确定对方能把事情做对,有些事情还是自己动手做更省事。
所以我想知道的是,对于像这样的硬件设备,你可以问它很多风险较低的问题,这没问题。但对于风险较高的事情,与其给出一个可能错误的答案,还不如自己查一下或者自己动手做更简单。
这会不会最终变成像 Google Home 那样的情况——你只是把它当个音箱用,问问天气,因为说错了也无所谓?但众所周知,只有当你能够信任它去回答那些风险较高的问题时,它才能真正发挥巨大作用。比如问:“嘿,艾米丽今晚的航班几点到?她住我这儿。她坐的是哪趟航班?”或者“航班延误了吗?”如果它答错了,那好吧,我就没法准备好去机场接她了。
这似乎确实是核心问题。进步会足够大吗?行业当然会宣称自己在模型幻觉方面取得了巨大进展。至于这是否是使用这些产品的人的真实体验,我认为这还有待商榷。但行业确实声称他们在这方面取得了很大进展。
另一个问题是,OpenAI 能否获取你的所有数据和个人背景信息,从而在智能音箱这类环境中让产品变得有用——在那里你可以对它随意说话?这又回到了它是否会有 Gmail 客户端的问题。Gmail 客户端将在哪里运行?你会让 OpenAI 登录你的 Gmail,这样当你询问航班信息时,它就能去查看吗?还是你只会默认在手机上打开 Gmail 并询问它,因为谷歌已经拥有这些数据?这似乎是一种竞争形态,实际上取决于你有多放心将你的所有数据交给不同的 AI 提供商。而我不确定 OpenAI 就是那个选择。
另一方面——显然这只是少数人的声音,并非 OpenAI 大多数用户——很多人对广告极为担忧。Anthropic 利用这一点,投放了反 AI 的超级碗广告。OpenAI 一直在应对这一负面影响,并试图表示:“嘿,别担心,我们不会影响回答内容,我们只是在回答下方投放类似的广告。”
但每当我使用 ChatGPT 时看到广告弹出,还是会感到不安。这让我觉得隐私性降低了。这感觉有点奇怪,我觉得每个人都会有这种不安感。即使它不影响实际回答,它仍然在提醒你,这个空间并非他们不试图向你推销东西的地方。总有人在试图向你推销东西。
目前 AI 实验室已经面临大量信任流失——如果说一开始有人信任它们的话。现在信任度正处于历史最低点。我们看到 AI 民粹主义正在抬头。正如《华尔街日报》近日报道的那样,AI 公司的 CEO 们收到了大量死亡威胁。反对声浪此起彼伏,就像你提到的大学生毕业典礼上对 AI 喝倒彩那样。眼下我们正目睹一场巨大的反弹,而我认为,现在这个时间点推出一款面向公众、"每个消费者都应该买"的 AI 硬件设备,实在有些奇怪。
这与 OpenAI 在苹果这起诉讼中面临的更大问题有关——当前 AI 行业基本上建立在未经授权的窃取之上。我之所以说"窃取"而不是"偷窃",是因为相关诉讼数量众多。我知道我们的观众希望我直接称之为偷窃,或许它确实就是偷窃。也许等到成千上万起诉讼尘埃落定之时,我们就能直接这么说了。但 AI 行业的这一面,感觉我们不得不刻意回避。
训练数据是否属于合理使用,这是一个价值万亿美元的问题,至今悬而未决。我们可以逐一审视那些诉讼——所有公司都必须为训练数据的使用进行辩护。如果它们输掉其中任何一场官司,整个行业的成本结构可能就会崩塌。
目前 AI 的输入是免费的。它们直接拿走了互联网上的内容,拿走了所有书籍。我们刚刚看到,由于一次黑客攻击,Suno 不得不基本承认它非法抓取了 YouTube 上的所有音乐。它根本没有获得授权,我认为 YouTube 对此肯定有意见,YouTube 上的艺术家们当然也有意见。
现在苹果表示,OpenAI 还窃取了它的硬件设计和制造工艺。这些同样未经授权——它们都是苹果的商业机密。OpenAI 雇佣了人员,策划了这场未经许可窃取所有这些东西的阴谋。
从文化层面看,这在我看来就是 AI 行业的一部分——他们认为可以不惜一切代价,尽快构建超级智能。一旦达成目标,一切都会被原谅。但公众似乎并不买账。艺术家、作家和音乐家显然不喜欢这样,现在苹果也不喜欢了。
感觉这场诉讼在很大程度上将成为这一问题的代言。未经许可就能拿走所有这些内容吗?我不确定 OpenAI 是否还有文化资本来反击。他们能说什么呢?
极具讽刺意味的是,整个行业之所以成为可能,完全依赖于它消耗的海量数据——无论这些数据是否合规。而 AI 实验室,乃至整个科技行业,当别人把那些信息用到别处时,却变得极其好诉。所以整件事相当讽刺。
AI 行业以“小圈子”著称,几乎是近亲繁殖。你在这里和同一批人共事,然后离开去别处。接着,同一家公司收购了你创立的初创公司,你又离开那家初创公司去别处,然后被你在另一家公司工作的老朋友通过人才收购招入麾下。很多行业都这样,但这个行业尤其如此。所以,是的,我认为这两方面都挺讽刺的。
有人会离开苹果,好吧,他们会带走一些商业机密。这极其恶劣吗?是的,其中一些指控确实离谱。但正如你提到的,这些公司最初能成功,很大程度上也是因为它们未经授权就拿走了大量东西。看着这一切上演,真是有趣的时刻。
所有 AI 公司都会非常恼火——比如说,当中国 AI 公司蒸馏它们的模型时。我只是在想,陪审团会怎么看,苹果会如何向陪审团陈述这个案子,然后说:“你偷走了所有音乐,不是吗?然后你还偷走了我们的制造设计。而当中国公司蒸馏 ChatGPT 时,你们不也生气吗?”
在你的报道过程中,AI 行业的人有没有意识到,对知识蒸馏感到愤怒,恰恰和艺术家、音乐家对自己所有作品被拿来训练模型时的感受一模一样?
我从未见过他们中任何人哪怕一次承认这一点,甚至暗示一下都没有,这太疯狂了。感觉这里存在一种脱节。他们对知识蒸馏如此愤怒,喊着“这违反我们的服务条款!”他们甚至不能说这是违法的,只是违反服务条款。他们给政府写公开信,甚至和死对头以及其他 AI 实验室联合起来对抗知识蒸馏。所以,我从未见过他们意识到这一点,这对我来说很有意思。
我知道很多人可能正在听这个,并且对我、对我们感到生气,因为我们没有声称苹果提起这起诉讼只是因为它在 AI 领域落后了。它在 Apple Intelligence 上失策了,花了很长时间才追上来,现在才正式推出。它不得不去找谷歌,蒸馏谷歌的 Gemini 模型,把它们放进新的私有云计算系统,以及所有它需要做的事情。但这些东西现在正在推出。
你是否觉得整个 AI 行业都在看着苹果因为落后而发难?因为我听到了很多这样的苗头,认为这只是苹果在推出 Apple Intelligence 并试图追赶之际,试图拖慢消费领域最大威胁的步伐。
这绝对可能是原因之一。所有人都知道,苹果在 AI 领域落后得出了名,多年来一直是热议话题和公司的尴尬事。所以,这绝对可能是原因之一。
但这起诉讼与有人为 Siri“窃取商业机密”无关。它关乎硬件,特别是硬件的“展示与说明”之类的事情。硬件是苹果做得极其出色的事情,而且几乎所有人都认为他们是这方面最棒的。
所以,我觉得有些人确实在这么想。如果苹果能通过任何方式削弱竞争对手、阻止其领先,这对他们来说是有利的。但这次诉讼具体涉及的是硬件方面的商业秘密,我认为这反而削弱了部分上述说法;这不像是在指控 OpenAI 窃取了 Siri AI 那些可能本就不多的秘密。
[笑] 他们就像在说:“请把我们知道的关于 Siri AI 的一切都拿走吧。那才真能拖垮你。” 苹果承受得起这种干扰,你之前在这段对话里也提到过。苹果可以雇一大群律师,让他们尽情发挥,想多难缠就多难缠。也许苹果的其他部门甚至都不会注意到这件事,因为他们正忙着制造下一代 iPhone。
苹果花了不知多少年、投入了大约 100 亿美元想造车,结果就是没成。苹果的其他部门甚至都没意识到这件事正在发生。他们只管继续出货手机和笔记本电脑,一切照常运转。
OpenAI 可不是那样的公司。OpenAI 可能连足够的高管来消化另一场干扰的成本都没有。他们的人一直在离职。OpenAI 的总裁 Greg Brockman,看起来几乎是在独自运营整个公司。
在你持续报道这件事的过程中,OpenAI 有没有意识到他们需要更多人手来做更多事情?因为我看这家公司,他们一直在流失人才,并且正如你所说,不断从一个方向转向另一个方向,现在他们又认定“我们需要和 Anthropic 竞争”。但看起来,他们无论如何都承受不起这种规模的干扰了。
他们承受不起。这就像是最终决战。实际上,后面还会有更多战役,但这是长期 AI 军备竞赛中的一场关键战役。今年是 IPO 之年。事情已经到了紧要关头。投资者正以前所未有的方式密切关注着这些公司。他们要求这些公司实现盈利,或者至少要有盈利的途径,而这是他们以前不太在意的事情,现在一切都到了关键时刻。他们现在承受不起任何干扰。
他们正在重组、调整架构、更换高管。当我问及此事时,他们会说:“嗯,这其实是一个大计划,我们只是分步实施而已。并不是说我们反复转向了无数次,我们始终朝着同一个目标前进,这些只是过程中的不同阶段。”但外界看来往往并非如此。
从某种程度上说,是的,格雷格·布罗克曼正在接管产品业务。这是因为AGI负责人菲吉·西莫因健康原因休假,随后又因健康问题永久离职。所以,这并非全是公司能控制的事情,但我确实认为,即便是普通员工中也出现了大量离职。他们需要招募更多人,必须加大招聘力度,而这需要更多资金,现在不是被任何事情分心的时候。
值得注意的是,有报道称OpenAI将推迟其IPO。就在你我交谈之际,SpaceX的股价已跌破其IPO发行价。这些大型IPO正面临某种风险。现在上市可能不是个好主意,尤其是当你没有足够的营收,或者更重要的利润来支撑时。你认为这场诉讼有没有可能成为压垮骆驼的最后一根稻草,促使OpenAI推迟IPO?
绝对有可能。报道显示,萨姆·奥尔特曼原本担心在Anthropic之前提交S-1表格。后来他们没有这么做。Anthropic抢先一步——这只是我的猜测——似乎Anthropic赢得那部分竞赛后,OpenAI就退缩了,说:“好吧,我们现在真的该做这件事吗?我们之前确实很想赢,但让我们看看实际情况吧。”这就是为什么它表示:“好吧,可能是明年初,而不是今年。”
那还是在苹果提起诉讼之前。所以现在,我确信OpenAI肯定在想:“好吧,我们是不是应该再推迟一些?尤其是我们根本不可能在IPO上击败Anthropic。”这似乎是个明智的想法。
如果你能推出一款出色的硬件设备,并以轰动且盈利的方式进入消费市场,那在IPO时就是个好故事。但如果你做不到这一点,因为你所做的一切最终都会陷入诉讼,或在走向诉讼的过程中陷入混乱的证据开示程序,那么也许这一切都只能先搁置一边。
这差不多是我的最后一个问题了。我感觉我在这个节目以及整个The Verge上的核心观点一直是:“好吧,我明白这对软件开发会如何运作,也明白这对企业级应用会如何运作,尽管那也还很不稳固。但消费者并不喜欢这些东西。” 有9亿免费ChatGPT用户这件事,除了说明“嗯,谷歌搜索变得很糟糕,而这东西会告诉你它爱你”之外,并不能反映其他任何东西。
但这并没有演变成“我们应该建更多数据中心”。也没有演变成“我们在文化上热爱AI产品”。它演变成了“把这些垃圾从我眼前拿开,我们会在毕业典礼上嘘你,我们会去市政厅会议说我们不想要数据中心。”
这里存在着巨大的鸿沟。你认为OpenAI能弥合这个鸿沟吗?因为它一直是一家追求宏大消费产品愿景、宏大消费市场的公司。如果它无法弥合这个鸿沟,那它就该关掉这块业务,去和Anthropic竞争企业级市场。我不知道会是哪一种情况。通过你所有的这些报道,你有没有感觉到OpenAI是否已经意识到消费者这一侧的争论有多么生死攸关?
我认为他们有一定程度的了解,但并非完全了解。他们会在演讲、研究论文和推文中提到这一点。他们有一些公关话术,会承认这是个问题,但我只是不认为他们意识到这道裂缝有多深。这个鸿沟太宽了。我看不到他们是唯一能弥合它的公司——如果它最终能被弥合的话。人们在AI方面对太多事情感到极度担忧,并且对此情绪非常激动。
这不仅仅是逻辑问题。我认为如果只是逻辑上的分歧——有些人从逻辑上认为它不好,有些人从逻辑上认为它好——他们或许还能弥合这种分歧。但一旦情绪介入,不,我认为他们无法完全弥合那道鸿沟。
对我来说,六年前我刚开始做 AI 记者时,我的朋友里没人真正知道或关心我写什么。那倒也没什么。这领域确实很小众,无所谓。大约四年前开始,这个话题会在酒吧、夜店里被提起。我走在街上,会有人问我关于 AI 的事,而他们并不知道我是 AI 记者。所以,当他们知道我是 AI 记者后,这种情况就更多了。但就是很随机地,这个话题会出现在人们面前,并且让他们变得极其情绪化。所以,是的,我不认为 OpenAI 会是那个弥合鸿沟的公司。
我们来看看硬件方面会怎样。我看到的很多人——TikTok 上的创作者、其他平台上的创作者——都在试图改变这里的叙事,他们说:“嘿,别被落下。学着用这些东西。” 这本身可能是有道理的,但他们很多时候的表述方式更像是:“哦,这东西学习曲线太高了。你必须学会怎么用它,否则就会被它甩在后面。”
学习曲线并没有那么高,这取决于你怎么用它,但你懂我的意思。有时候确实很高,是的,如果你在用 Codex 做各种高强度的氛围编程,那当然高。但如果你只是跟聊天机器人聊聊天,那学习曲线真的不高。
我个人认为,所有这些算法推送的社交视频内容都是一场伪草根营销活动。这只是我的看法。这是个阴谋论,我们暂且不谈。我倒是想向你推荐一个视频系列,叫“海登在夜店”,就是你去夜店,跟人们聊聊 AI。
[笑] 我很乐意正式地去做这件事,而不是像现在这样总是非正式地碰上。
我已经很久没去过夜店了。我年轻的时候,夜店可不是这样的。时代变了。
海登,这次聊得非常愉快。我强烈预感,今年剩下的时间里你还会再上这个节目,继续谈这起诉讼和 OpenAI 的事。我们得尽快再请你来。
期待之至。谢谢。
有问题或意见?请发送邮件至 decoder@theverge.com。我们真的会阅读每一封邮件!
Today on Decoder, I’m talking with Hayden Field, The Verge’s senior AI reporter, about the major trade secrets lawsuit between Apple and OpenAI and what this tells us about OpenAI’s future.
By now I’m sure most Decoder listeners are familiar with Apple’s allegations in this case. The company says a number of ex-Apple employees at OpenAI targeted Apple’s trade secrets in job interviews and even downloaded files from Apple’s servers related to hardware manufacturing. The claims are shocking, although OpenAI denies them, and we have yet to see the company’s legal response.
Hayden has been covering OpenAI for years now, and she also just talked to a bunch of lawyers who specialize in intellectual property and trade secret law to get a sense of how serious this case is. The answer is not entirely positive for OpenAI, which spent $6.5 billion to acquire legendary Apple designer Jony Ive’s AI hardware startup, io Products, back in 2025.
As you’ll hear Hayden explain, Apple is a notoriously tenacious litigant, and we have lots of tech industry history to back up that reputation. But those cases were against companies at the scale of Microsoft and Samsung, giant corporations that could afford the cost and distraction of this kind of litigation.
OpenAI isn’t yet that kind of company. It’s still burning cash and a little strategically adrift as it tries to compete in both enterprise and consumer markets, and it’s turning over executives on a regular basis. So Hayden and I spent a lot of time talking about what this case might mean for a company that defined the early part of the AI boom, but which lately always seems to be in a crisis of one sort or another.
Okay: Verge senior AI reporter Hayden Field on the Apple-OpenAI lawsuit, and how the AI industry might change because of it. Here we go.
Hayden Field, you’re the senior AI reporter for The Verge. Welcome back to Decoder.
Thanks. Great to be here.
Once again, we’re going to talk about OpenAI, Hayden.
Once again, it never stops.
We’re going to talk about whether OpenAI can maintain focus at a time of great distraction, which seems to be a theme with that company. Although this time, it’s not their own distraction. It’s a distraction from out in the world, maybe caused by some of their own behavior.
Of course, we’re talking about the Apple trade secret lawsuit. It feels like a big deal. Apple is accusing OpenAI of stealing all kinds of information from that company, even in job interviews — asking people in job interviews to tell them what they know about Apple products. Tell us briefly about this lawsuit and what’s going on here.
Definitely. So, it is crazy. I wrote a piece last week titled “Sam Altman didn’t need another lawsuit,” and that’s really the situation here. It seems pretty egregious from some of the claims that Apple is making. A former Apple employee was allegedly soliciting trade secrets over interviews, asking people to maybe bring pieces of hardware outside the Apple office to OpenAI to do a “show and tell” in interviews, things like that.
There’s a lot of allegations being made that are pretty crazy. But then when I actually interviewed a bunch of experts last week, they said they’ve seen all of this before, a lot. It’s just they haven’t seen all of it in one case with two key players.
Some of this stuff is pretty egregious, if it turns out to be true. But also, to the experts in the space, they’re saying, “Oh, we see this all the time with trade secrets cases. It’s not crazy. What’s really crazy is that it’s all in one case and that it’s two really large companies.”
It’s really interesting. It’s just a pattern, maybe not worth diving into at great length, but in the ’90s, Apple pushed copyright law to the limit. They sued Microsoft for allegedly infringing on how the Mac looked and felt with Windows. I will point out that Windows still exists, Microsoft is fine.
And then in the 2000s and early 2010s, Apple pushed patent law to the limit, suing Samsung for allegedly violating a bunch of patents with Android and Galaxy phones because Apple felt that they had protected the iPhone. Samsung eventually paid about $1 billion in damages and those cases just went around in circles forever.
But I’ll point out that Android still exists, Samsung is doing just fine. Apple may have redefined patent law around 3G and 4G licensing, but fundamentally, it did not actually succeed in stopping its competitors with intellectual property lawsuits.
So, you go from copyright to patent and now trade secrets, which is the third big intellectual property, and Apple’s doing it again with OpenAI. My big question is whether OpenAI can just pay the money the way that Microsoft and Samsung were able to just pay the money. If you look at Samsung, spending $1 billion in penalty fees to get Galaxy phones off the ground and be a major competitor in mobile is maybe the easiest and best investment that company ever made.
But I don’t know if OpenAI is the same company at the same moment. It certainly seems like all of the Apple employees that went over to OpenAI maybe pushed the limit too much, at least from the story that Apple’s telling us right now.
It’s pretty crazy reading the wildest claims from the lawsuit and also the messages that were being exchanged. One of them allegedly accessed Apple confidential information on a device and then said, “LOL! So funny that I can access this.” There’s a lot there.
I’ve been wondering the same thing. It really depends on Apple’s willingness to be petty or be super logical here. When I interviewed a bunch of lawyers about this, they said Apple is a notoriously tenacious litigant, as we’ve seen in those past cases. But also, they really will take things far. They have a lot of money and they’re willing to use it. It really just depends on how mad they are.
I think that they would potentially settle if OpenAI paid them enough, but will OpenAI be able to do that? They’re already having a really tough year. They’re about to IPO, all the investors are looking at their bottom line, wondering when they’re going to turn a profit. It’s a tough time to be shelling out a ton of money to Apple. We’ll have to see what they do.
I think it’s going to go to trial and be a very drawn-out years-long thing. That’s what a lot of the lawyers I interviewed said as well. But I could see a world in which they just figure out a way to get all this money together and pay and maybe Apple accepts it.
In the Samsung patent cases, they lasted forever. In the middle of it, Tim Cook was at a Code Conference and I asked him, “What are you doing? Is this accomplishing anything?” And he looked at me very seriously and he said, “Well, we think the fees for licensing standard essential patents are out of whack.” And I was like, “That is not the reason you’re doing any of this. That’s not the point, the point is that you were mad and you wanted to run this lawsuit to ground and you’re going to run this lawsuit to ground no matter what the nominal reason is.”
That feels like a big piece of the puzzle here. OpenAI has taken a bunch of Apple employees. Famously Jony Ive, Apple’s most famous designer, is working with OpenAI on hardware devices. Tang Tan, who is OpenAI’s chief hardware officer. He had a reputation at Apple for “playing fast and loose,” that’s from Mark Gurman at Bloomberg.
He seems like the central character in the story, not so much Jony Ive, but Tang Tan. And that piece where it’s a bunch of ex-Apple people maybe playing too fast and loose almost feels like it’s personal, almost feels like there’s a lot of emotions involved. Talk about Tang Tan and talk about what all these folks are alleged to have done that feels like it’s a little more personal than your standard big corporate dispute.
On what you said about talking to Tim Cook, I truly believe people underestimate companies’ pettiness. We’ve seen it in recent years, like with the OpenAI-Elon Musk lawsuit. There are so many things that could have been avoided if companies weren’t trying to just be petty, which is fine. They can do that, they have enough money to do that. But people underestimate how much that plays a part sometimes in these things.
As for Tang Tan, yeah, he worked at Apple for 24 years. He was the VP of the Apple Watch. He left in 2024 to work on Jony Ive’s hardware company, io Products, which obviously OpenAI ended up purchasing.
OpenAI then appointed him as chief hardware officer. So he had a C-suite role at OpenAI, but this is after 24 years at Apple. According to Apple’s lawsuit claims, he was the mastermind of a lot of this trade secrets gathering, regarding interviews that he would conduct and ask about secret code-named projects. And then the person being interviewed would feel like maybe they were able to tell him stuff because, oh, well, why would he know the name of this project or these details about it if he wasn’t involved also? So, they felt they could speak more freely, allegedly.
There were other times that he asked, “Hey, why don’t you bring some batteries or other pieces of hardware outside the Apple office that you’ve worked on in the past so that we can do a show and tell?” It’s very interesting to go through these claims. It just seemed like he really was not worried about what was going to come of this, if these claims are true.
The other major character here is Jony Ive, Apple’s very famous ex-designer who worked with Steve Jobs and all of those famous projects. I would say Ive reportedly has a very testy relationship with Apple.
Now that he’s left, he’s certainly made that clear in various interviews and statements. And he’s working with Sam Altman on new devices. He made a video with Sam Altman, which I can only describe as a romantic coffee date.
Absolutely.
In which they described the phone and the laptop as legacy devices.
It was instantly memed. I have to say the memes were amazing.
It’s very good. And they shut down half of San Francisco to make it. I thought Ive calling the phone and the laptop legacy devices in that video was just a straight shot at Apple. These are the things that he made, that he revolutionized in many, many ways. And he’s like, “These are over now.”
You can say a lot about whether that’s true, whether that can come true, and I want to talk about that in a little bit. But it’s notable to me that Jony Ive is not mentioned in this lawsuit. There’s a line Apple wouldn’t cross, and it’s mentioning the fact that it’s Jony Ive who went to OpenAI and started hiring all of his old designers, who sold a big portion of his company to OpenAI. And now, they’re trying to make hardware and allegedly stealing trade secrets. But Jony Ive, he’s not part of it. We’re just going to leave his name out of it.
Is that because he’s too famous? Is that because it’s too nuclear? Is it because he’s still friends with Laurene Powell Jobs? What do you make of that?
I think it may be more that he knew how to cover his tracks. It could be that he’s too powerful or any of the things you said, but I really feel like it’s just that he has other people doing his dirty work for him. The other stuff in this lawsuit is pretty egregious, and they’ve found it all. So if they found some really intense stuff on Jony Ive, would they really not put it in the lawsuit? There’s some bad blood between them. I just feel like they would put it in there.
This is just speculation, but when you’re in a powerful enough role, you don’t have to do all this stuff yourself. You have other people doing it for you and you also know how to cover your tracks. So, to me, maybe they just didn’t have anything on him.
Yeah, or he’s distracted designing Ferraris no one likes or whatever it is he’s doing, but he’s not in this. I do think that’s notable because the reason Sam Altman would hire Jony Ive is to make a credible claim to the next device. Whatever that category of AI-powered device is going to be, you hire the most famous designer to say, “I made the iPhone and I’m going to make this thing.”
And then it seems like OpenAI also wanted a bunch of Apple’s know-how in order to make that thing. What do you make of that, that OpenAI needed hardware know-how from Apple? Or at least Apple alleges that OpenAI needed hardware know-how from Apple to make whatever the next thing is?
To some extent, whether you’re talking about trade secrets or just regular legal aboveboard know-how from hiring their employees, of course they needed it. They’ve seen a ton of hardware in AI crash and burn and they’re trying to not become the next Humane AI Pin. It can be incredibly embarrassing to make an AI hardware device in this day and age. Chances are you’re probably going to fail miserably and become a laughingstock. And so, if you want to avoid that, of course you need to get the best people on your side.
Hardware is hard, as everyone says, so of course they want to get experts in there that have experience actually being incredibly successful at hardware. That’s not surprising to me and I think it’s a smart decision. OpenAI has demonstrated in the past — and this is typical of every tech company, it’s not just OpenAI— that sometimes they’re a little bit disconnected from what the consumer actually wants.
You roll something out, no one really wants it. Okay. With software that’s a little easier to pivot from, as we’ve seen from OpenAI deleting its side quests and stuff. They’re like, “Oh, we’re not doing this anymore. We’re not doing this anymore. We’re rolling this into that.” With hardware, you can’t do that. You’d better have it right from the get-go.
I’m glad you said consumer, because that brings me to what I think is the heart of the matter for OpenAI, which is that it was going to be the consumer AI company, Anthropic famously focuses on enterprise and it’s been lucrative for them; maybe that’s where all of the money is. But OpenAI was going to take down Google, maybe through Bing, maybe through its own app. It was going to be a consumer company.
The only reason you would make hardware is to go attack the consumer, to go attack the iPhone. So, OpenAI’s ambitions, at least as I could see them, were first to take market share from Google, which is one of the most lucrative businesses in the history of the world, and then to go attack the iPhone, which is the most famous and culturally dominant tech product that has ever existed.
I don’t know how it can do either of those things. It doesn’t seem like consumers even want these products or want AI at this scale. I don’t know where that revenue comes from in consumer, and the college students are all booing AI at graduation. Does OpenAI just need to pivot to enterprise? Does it need to spend its time spinning its wheels at consumer hardware devices before it figures out its business?
OpenAI has pivoted so many times internally. Its most recent pivot was that it was going to focus on its key revenue drivers, which are enterprise and coding. They’ve said that a million times. OpenAI folded Sora, Atlas, the sexting feature. It feels like years ago, but it’s crazy that that was in the past year.
They’ve done a lot of this, and so, everyone thought, “Okay, they’re going for it, they’re doing enterprise,” because that’s the tried-and-true money maker here, especially in a year that they’re going to IPO. To be fair, yes, they had already planned to do the hardware device before this whole pivot and cutting out the side quests. I guess they can’t afford to cut out this potential side quest because they’ve already spent a lot on it, it’s happening no matter what. So now they’re just trying to make it as successful as it can.
I’ve only been on the AI beat for six years. That is a long time in AI, and maybe I’m missing something, but I just do not see a market when everyone is buying an OpenAI hardware device. If I’m wrong, we can come back on here and laugh at my clip from July 2026. A smart speaker, okay, maybe people buy it just to be a smart speaker like Google Home, that’s what I use mine for.
I think it’s going to be one of those things that a few people really, really, really love and that the general public probably does not adopt because it’s difficult to get them to adopt anything unless all their friends have it. There’s a social aspect. You’re going to be a laughingstock with green text if you don’t have an iPhone. No hate to Android people, I’m just saying sometimes you don’t get added to the group text or something. It’s sad and it’s not right, but there’s got to be some public buy-in if you’re going to be adopting a new piece of hardware a lot of the time.
We founded The Verge at the start of the smartphone boom. Just thinking back on it, it was never a question of whether people would adopt smartphones. That was obvious. It was obvious to us, maybe it wasn’t obvious to everyone, but we were in it and the only thing that wasn’t obvious was who would win.
After a while it became obvious that Apple would win. How many iPhone killers did we cover 15 years ago? And they all came and went, and they looked like new operating systems. Windows Phone existed. The Palm Pre ran an operating system called webOS. Android was ascendant. Nokia tried to do its own thing and then they pivoted to Windows Phone.
At the end of the day, we ended up with iOS and Android and everything else fell into line. Maybe the whole tech industry hates this and they hate paying the App Store fees, but this feels like that same kind of moment where OpenAI is going to put out a hardware device, it’s going to run its own operating system. Elon Musk is fulminating about Starlink and xAI putting out their own phone, maybe running their own operating system, apparently the run-up to IPO, they were showing it to people.
I just have really dumb questions that, to me, mirror the same questions we asked about new smartphone operating systems 15 years ago. Are you going to make a Gmail client? Is it going to have Instagram? Because if you don’t have those things, you don’t actually get to replace the phone. And if you don’t replace the phone, well, I’ve got the beta of iOS 27 on my iPhone right here and Siri AI is definitely good enough to just get through the day.
That’s what really gets me here, because the smartphone replaced the dumb phone. So okay, you’ve got to have something way better than the smartphone. Are you going to? Doubtful, but we’ll see.
But the other thing that gets me here is how many devices OpenAI is reportedly working on. Reportedly it’s five hardware devices, and the screenless smart speaker is the first one. How can they come up with five? That’s what I want to know. What are the five? And shouldn’t you be putting all your resources toward one? It’s already hard to make one. I’m a little dubious of five successful AI hardware devices. Even one is a long shot, but five is a little bit crazy.
I’m sure there’s a lot and some will come and go. The thing that I think about, particularly when it comes to Jony Ive and his approach to hardware — which is familiar at this point, we’ve watched him launch a lot of hardware at Apple — is Ive tends to focus on new input methods for new devices.
The iPod had the click wheel and that really defined the form and function of the iPod. The iPhone has a multitouch screen that really defined the form and function of the iPhone and then the iPad. If you’ll remember the Apple Watch, they launched it with great fanfare about the scroll wheel on the side. I don’t know if that came to pass, but they really thought that was a big deal when they launched the Apple Watch.
This is Ive’s go-to path: “We find a new input mechanism, we’re going to define the product around it.” I know exactly what that should be for AI — it’s a natural language input. You can just tell the computer whatever you want and it’ll figure it out and it’ll go off and do it for you.
But we’re however many years into the AI revolution?Three or four years into this version of it? And it doesn’t quite work yet. I think this is why the AI devices keep failing, as you keep mentioning. Because you can talk to a Humane Pin all day and all night and it will pretend to understand you and then not actually be able to do the things on the other side of it.
That feels like the problem in AI right now. There’s these flashes of brilliance and then there’s these limits of capability. Where do you think we are in that process of figuring out exactly what it can and can’t do? And how do you think that limits a hardware device?
We’re still in the midst of it. I personally use AI all the time, obviously in my personal life to test things out, see what it can do. Month to month, I see how it’s improved, at least from my own personal experience, which is all anecdotal.
But just the other day I was trying to look up a lot of addresses of companies because I wanted to map something out. I would say maybe 60 percent of the ones I looked up were completely wrong, just wrong. It made me a whole table. And it was like, “Yeah, don’t worry, all these are definitely correct.” And then 60 percent wrong. I’m like, “A lot of these are wrong.” They’re like, “Oh, sorry, my mistake.”
Yeah, my favorite.
It’s like, “Oh yeah, you’re right. You caught me.”
It’s right about a lot of stuff a lot of the time. But as we all know from when we were an intern back in the day, or if we’ve ever had an intern, there are certain things that are just easier to do yourself if you’re not 100 percent sure the person is going to get it right.
So what I wonder is, for a hardware device like this, there are a lot of low-stakes questions you can ask it and it’ll be fine. But for higher-stakes things, it’s simpler to just look it up yourself or do it yourself even than put it out there when it’s going to be a potentially wrong answer.
Is this going to end up like a Google Home situation, where you’re just using it as a speaker and you’re asking it what the weather is because it doesn’t really matter that much if it’s wrong? But as we know, it can only really be super useful if you can trust it with those higher-stakes types of questions. And so, saying, “Hey, what time does Emily’s flight get in tonight because she’s staying with me? What flight is she on?” Or, “Is it delayed?” If it gets that wrong, okay, I’m not going to be ready for her to come in from the airport.
This does seem like the core problem. Will there be enough progress? Certainly the industry will claim that it’s made great progress on hallucinations. Whether that’s the lived experience of people who use these products, I think that’s up for debate. But the industry claims they’ve made great progress here.
The other piece is, will OpenAI have access to all of your data and personal context to make this useful in an environment with a smart speaker, where you can just say whatever you want to it? That, again, brings you back to whether it’s going to have a Gmail client. Where’s the Gmail client going to run? Are you going to let OpenAI log into your Gmail, so when you ask for flight information, it can go look at it? Or are you just going to default to opening Gmail on your phone and asking it because Google already has it? This seems like a version of competition that really depends on how comfortable you are handing over all of your data to various AI providers. And I’m just not sure OpenAI is that one.
The other part of this is that — and obviously this is a vocal minority here, it’s not like the majority of OpenAI users — a lot of people are extremely worried about ads. Anthropic capitalized on that with its Super Bowl ads that were anti-AI ads. OpenAI has been dealing with that fallout and just trying to say, “Hey, don’t worry, we’re not going to influence answers, we’re just going to raise a similar ad underneath the answer.”
But it can be unnerving, whenever I use ChatGPT and an ad surfaces. It just makes me feel like it’s less private. It’s just a little bit strange and I feel like everyone gets that unnerved feeling. Even if it’s not affecting the actual answer, it’s still reminding you that it’s not a space where they’re not trying to sell you something. Someone is always trying to sell you something.
There’s already a lot of loss of trust right now in AI labs, if anyone trusted them to begin with. Right now, it’s at an all-time low. We’re seeing the rise of AI populism. We’re seeing lots of death threats against AI CEOs, as The Wall Street Journal covered the other day. There’s a lot of backlash, like you were saying about college kids booing AI at graduations. We’re seeing a huge backlash right now, and I just think it’s a strange time to be releasing a public-facing, “every consumer should buy this” AI hardware device.
That connects to a bigger issue for OpenAI in this Apple lawsuit, which is that the AI industry right now is basically built on unauthorized taking. I’m saying that instead of theft because there are a lot of lawsuits. I know our audience wants me to just call it theft and maybe it is just theft. And maybe we can just say that when we get to the end of the road of 10,000 different lawsuits. But that part of the AI industry feels like we have to studiously ignore it.
There’s a trillion-dollar question over whether training data is fair use that is as of yet unresolved. We can just go down the list of lawsuits where all of these companies have to defend themselves about training data. And if they lose any one of them, maybe this whole industry cost structure falls apart.
The inputs to AI right now are free. They just took the internet and they took all the books. We just saw, as a result of a hack, that Suno had to basically admit that it had scraped all of the music off of YouTube illegally. It just didn’t have permission, and I think YouTube feels a certain way about that and certainly the artists on YouTube feel a certain way about that.
Here’s Apple saying OpenAI also took its hardware designs and its manufacturing know-how. It didn’t have permission for that either, those are Apple’s secrets. OpenAI hired people and they concocted this scheme to take all of that stuff without permission.
Again, culturally, this just feels like part of the AI industry to me, that they think they can just take everything to build super intelligence as fast as they can. And once they get there, all will be forgiven. But the public doesn’t seem to like it. Certainly, artists and writers and musicians don’t seem to like it, and now Apple doesn’t like it.
It feels like this lawsuit is going to be a proxy for that as much as anything else. Can you take all this stuff without permission? I’m not sure OpenAI has the cultural capital to fight back. What are they going to say?
It’s really ironic that this whole industry is only possible because of the amount of data it’s consumed, whether aboveboard or not. And AI labs, as well as just tech companies in general, get so litigious when others take that information elsewhere. So this whole thing is pretty ironic.
The AI industry is really famously cliquish. It’s almost incestuous, it’s like you work with the same people here, then you depart and go elsewhere. Then that same company buys your start-up that you created and then you leave that start-up to go somewhere else and you get acqui-hired by your old friend that worked at this other place. A lot of industries are like that, but this industry specifically is so much like that. And so yeah, I think it’s a bit ironic on both counts.
People are going to leave Apple, and okay, they’re going to take some trade secrets. Was this incredibly egregious? Yes, some of these claims are crazy. But also, like you were mentioning, a lot of the reason that any of these companies are even successful in the first place is because of all the unauthorized taking they’ve done. It’s an interesting time to be watching all this play out.
All the AI companies get really mad when, I don’t know, Chinese AI companies distill their models. I’m just thinking about a jury and how Apple will play this case to a jury and be like, “You stole all the music, didn’t you? And then you also stole our manufacturing designs. And don’t you get mad when the Chinese companies distill ChatGPT?”
In the course of your reporting, does it ever occur to the people in the AI industry that being mad about distillation is the exact same feeling that artists and musicians have about all of their work being put into the model in the first place?
I haven’t seen any of them even acknowledge that or at least even hint toward it one time, which is crazy. It just feels like there’s a disconnect there. They get so, so angry about distillation, they’re saying, “It’s against our terms of service!” They can’t even say it’s illegal, just against the terms of service. They write open letters to the government. They even come together with their mortal enemies and other AI labs to unite against distillation. So, it’s just interesting to me thatI’ve never seen that occur to them.
I know a lot of people are probably listening to this and that they’re mad at me, and mad at us, for not making the claim that Apple is just filing this lawsuit because it was behind in AI. It missed with Apple Intelligence, it took a long time to catch back up, it’s only shipping it now. It had to go to Google and distill Google’s Gemini models and put them in its new Private Cloud Compute system, and all the stuff that it needed to do. But that is shipping now.
Do you feel like the broader AI industry is just looking at Apple lashing out at being late? Because I have heard a lot of glimmers of that, that this is just Apple trying to slow down the biggest threat in consumer as it launches Apple Intelligence and tries to catch back up.
That could be definitely part of it. Everyone knows Apple has been famously incredibly behind in AI and it’s been a big topic of conversation and an embarrassment for the company for years. So, that could definitely be part of it.
But this lawsuit is not about someone “stealing trade secrets” for Siri. It’s about hardware, specifically “show and tell” of hardware, things like that. Hardware is something that Apple does amazingly and they are pretty much by all accounts the best at.
So, I think some people are thinking that. It’s a benefit for Apple if they’re able to kneecap a competitor from pulling ahead in any way. But the fact that this is a trade secrets lawsuit about hardware specifically I think has rolled back a few of those claims; it’s not like they’re trying to claim that OpenAI is stealing Siri AI’s secrets that it may not have many of.
[Laughs] They’re like, “Please take everything we know about Siri AI. That will truly kneecap you.” Apple can afford the distraction, you’ve mentioned this already in this conversation. Apple can just hire a bunch of lawyers and set them off to be as tenacious as they want. Maybe the rest of Apple won’t even notice that that’s happening, because they’re busy making the next iPhone.
Apple spent however many years and like $10 billion trying to make a car and it just didn’t work. It didn’t even occur to the rest of Apple that was happening. They just kept shipping phones and laptops and everything was just fine.
OpenAI is not that company. OpenAI might not even have enough executives to absorb the cost of another distraction. They keep leaving. Greg Brockman, who’s the president of OpenAI, seems to be running the entire company himself.
As you’ve been reporting on the story, has it occurred to OpenAI that they’re going to need more people to do more things? Because I look at that company and they just keep shedding all of their talent and pivoting, as you said, from one thing to the other and they’ve landed on “we need to compete with Anthropic.” But it doesn’t seem like they can afford distraction at this scale in any way, shape, or form.
They can’t. This is like the final battle. Actually, there are going to be a lot more battles, but this is a key battle in the longstanding AI arms race. This is the IPO year. Things are coming to a head. Investors are looking very closely at these companies in a way they haven’t before. They’re looking for them to turn a profit, or at least have a way to turn a profit in a way that they have not really paid too much attention to before, and it’s all coming down to the line. They cannot afford a distraction right now.
They are reshuffling, restructuring, changing executives around. When I’ve asked them about this, they will say, “Well, it’s all been one large plan that we’ve just been doing in little pieces. It’s not like we’ve pivoted a million times, we’ve always been going towards one thing and these are just different steps along that journey.” But it doesn’t always read that way to the general public.
In some ways, yes, Greg Brockman is taking control of product. That’s because Fidji Simo, the AGI chief, took leave because of her health and then she left permanently because of her health. So, it’s not all things that the company could control, but I do think there have been a lot of defections even from the rank-and-file staff. They’ve got to get more people. They’ve got to hire more, which will require even more money, and it’s not the time to get distracted by anything.
It’s worth noting that there’s some reporting saying OpenAI will delay its IPO. And as you and I are sitting here speaking, SpaceX’s stock price has dipped below its IPO number. These mega IPOs, they’re in some kind of danger. It might not be a good idea to do this, especially if you don’t have the revenue or, more importantly, the profits to back it up. Do you think there’s some risk that this lawsuit is the straw that pushes OpenAI to delay its IPO?
Absolutely. Reporting showed that Sam Altman was originally worried about filing its Form S-1 before Anthropic. Then it didn’t do that. Anthropic did it first and then — and this is me just speculating — it seems like after Anthropic won that part of the race, OpenAI backed off and said, “Okay, well, should we actually do this right now? We were trying really hard to win, but let’s look at the facts here.” That’s why it said, “Okay, it might be early next year, it might not be this year.”
That was before Apple’s lawsuit dropped. So now of course I’m sure OpenAI is thinking, “Okay, should we delay it even more? Especially because there’s no way we’re going to beat Anthropic to an IPO.” It seems like that might be a smart idea.
If you can put out a great hardware device and make some big splashy profitable entry into the consumer market, that’s a good story to tell in an IPO. But if you can’t do that because everything you do is going to end up in litigation or in messy discovery on the way to litigation, maybe that all just has to be set to the side.
That’s pretty much my last question here. I feel like my thesis on this show and on The Verge generally has been, “Okay, I understand how this is going to work for software development and I understand how this is going to work in enterprise, although even that’s pretty shaky still. But consumers don’t love this stuff.” That there are 900 million free ChatGPT users is not a reflection of anything other than, well, Google Search got really bad and this thing will tell you it loves you.
But it hasn’t turned into, “We should build more data centers.” It hasn’t turned into, “We culturally love the products of AI.” It’s turned into, “Get this slop out of my face and we’re going to boo you at graduation and we’re going to come to the town board meetings and say we don’t want data centers.”
There’s some huge gap there. Do you think OpenAI can close that gap? Because it has always been the company chasing the big consumer product vision, the big consumer market. If it can’t close the gap, it should just shut it down and compete with Anthropic for enterprise uses. I don’t know which one it’s going to be. Do you have a sense through all this reporting if OpenAI has realized how existential the consumer side of the debate is?
I think they know to some extent, but not the full extent. tThey will get at it in speeches, in research papers, in tweets. They have some PR lines where they will acknowledge that this is a thing, but I just don’t think they realize how deeply this crevice goes. The gap is so wide. I don’t see them being the only company to close it — if it’s ever going to be closed. People are extremely worried about so many things when it comes to AI and they’re very emotional about it.
It’s not just logic. I think if it was just a logical gap — some people logically think it’s not good, some people logically think it is good — I think they could probably close that. But when emotions come into play, no, I don’t think that they will be able to fully bridge that gap.
For me, when I became an AI reporter six years ago, none of my friends really knew or cared what I wrote about. That was fine. It was really niche, whatever. Starting four years ago, it would come up at the bar, at the club. I’ll be walking down the street, someone’s asking me about it, and they don’t know I’m an AI reporter. So, when they learn I’m an AI reporter, it happens even more. But just randomly, this is coming up for people and making them extremely emotional. So yeah, I just don’t think OpenAI is going to be the company to bridge that gap.
We’ll see what happens with the hardware of it all. A lot of people I’ve been seeing — TikTok creators, creators on other platforms — are trying to change the narrative here, saying, “Hey, don’t get left behind. Learn how to use this stuff.” And it’s something that can be valid, but the way that they’re framing it a lot of the time is more, “Oh, it’s such a high learning curve. You must learn how to use it before it leaves you behind.”
It’s not that high of a learning curve, it depends on how you’re using it, but you know what I mean. Sometimes it is, yeah, if you’re using Codex and doing all sorts of intense vibe coding, yes, of course it is. But if you’re just chatting with a chatbot, it’s really not a high learning curve.
I personally believe all that algorithmic social video content is an astroturfing campaign. That’s just me. That’s a conspiracy theory, we’re going to leave that aside. I do want to pitch you a video series called “Hayden in the Club,” where you just go to the club and talk to people about AI.
[Laughs] I would love to do that officially instead of just having it happen to me unofficially all the time.
It’s been a long time since I was in the club. That was not what the club was like when I was younger. Things have changed.
Hayden, this has been great. I have a strong suspicion you’re going to be back on this show talking about this lawsuit and OpenAI for the rest of this year. We’re going to have to have you back very soon.
Can’t wait. Thanks.
Questions or comments? Hit us up at decoder@theverge.com. We really do read every email!