内容来源：https://www.theverge.com/ai-artificial-intelligence/929129/sam-altman-testimony-elon-musk-openai-trial

内容总结：

经过两周听取各类证人指证“他是个爱说谎的蛇”，陪审团终于听到了这条“蛇”本人的陈述——萨姆·奥尔特曼。在证词结尾，他的律师威廉·萨维特问他，被指控窃取一家慈善机构是什么感受。奥尔特曼在证人席上表现不俗，但这可能并不足以扭转局面。埃隆·马斯克或许已对这位OpenAI首席执行官造成了更深远的声誉损害。

“我们通过大量辛勤工作创建了这个极为庞大的慈善机构，我同意你无法窃取它，”奥尔特曼说，“但马斯克先生确实试图扼杀它，我想，还不止一次。”奥尔特曼完全进入了“圣路易斯好孩子”模式，对被卷入这一切表现得一脸无辜。当他抱着一摞证据文件走下证人席时，甚至有点像个小学生。在直接陈述的初期他显得紧张，但很快就进入了状态。总体而言，他的证词显得可信——有时陪审团似乎还对他颇有好感。

整个庭审过程中，我很难完全揣摩陪审团的想法，因为我本人对出庭的这些人物过于熟悉。我听到过一些极其大胆的伪证，比如埃隆·马斯克当众声称自己从不发脾气（随后他在交叉质询中当场发飙）。又比如他几个孩子的母亲希冯·齐里斯告诉我们，她不知道马斯克在创办xAI——这似乎与她自己的短信记录直接矛盾。再比如格里高利·“什么能让我赚到10亿美元？”·布罗克曼信誓旦旦说自己一心只为使命。我当然相信奥尔特曼并不可信——毕竟《纽约客》曾用超过1.7万字的篇幅揭露他有多爱撒谎。但和马斯克不同的是，有同期文件可以佐证奥尔特曼的说法，至少大部分如此。

奥尔特曼作证称，他对马斯克坚持要控制权感到不安，不仅因为马斯克不像其他人那样深度参与，更因为OpenAI成立的本意就是不让任何一个人掌控通用人工智能。他在Y Combinator担任总裁时见过太多控制权之争——没人愿意在顺风顺水时放弃权力。通过超级投票权之类的架构，创始人可以永远保持控制。有趣的是，奥尔特曼举的例子并不是最著名的那个（Meta的马克·扎克伯格），而是马斯克本人和SpaceX。当奥尔特曼询问马斯克关于OpenAI的继任计划时，得到了一个“令人毛骨悚然”的回答：如果马斯克去世，他说，“我没怎么想过这个问题，但也许控制权应该交给我的孩子们。”

我确实看到了一封2017年奥尔特曼发给齐里斯的邮件，其中写道：“我担心控制权问题。我认为任何人都不应控制世界上第一个通用人工智能——事实上，我们创办OpenAI正是为了防止这种情况发生。”他接着表示，自己不介意短期控制权，并愿意接受“创造性的结构”——我的理解是，为了安抚马斯克，奥尔特曼愿意在达到某些公司发展里程碑之前给予他控制权。

奥尔特曼在证人席上说：“我认为他想要的是长期控制权，如果我们当初同意了他想要的结构，他就能实现这一点。”这听起来基本正确。在后续萨姆·特勒的证词录像中，我们得知马斯克不再投资任何他不掌控的公司。这也符合马斯克长期以来的执念——确保自己不会像在PayPal被踢出局那样从自己的公司被赶走。

马斯克还试图将奥尔特曼招募到特斯拉。我们看到了奥尔特曼与特勒之间的短信，特勒告诉奥尔特曼，马斯克无论如何都要加强特斯拉的人工智能，并希望奥尔特曼、布罗克曼和伊利亚·苏茨克维最终能加入。“我从那里面读到了一种模糊的、轻微的威胁——无论有没有你们，他都会在特斯拉内部搞这个，”奥尔特曼说。但他认为特斯拉首先是一家汽车公司——让它收购OpenAI会背叛OpenAI的使命。

后来在特勒的证词中，我们看到2018年2月4日凌晨12点40分特勒发给齐里斯的短信：“我不希望OpenAI在没有埃隆的情况下继续下去，”他写道，“我更想通过挖走其领导层来搞垮它。”

当马斯克停止季度捐款时，OpenAI的资金“极其紧张”，“现金跑道极短”。OpenAI确实有其他捐赠者，但没有一个起诉它或加入马斯克的诉讼（庭审中未被提及的一个捐赠者是阿拉梅达研究公司，它属于现在因欺诈和洗钱入狱的萨姆·班克曼-弗里德）。马斯克从董事会辞职意味着“人们怀疑他是不是要对我们实施报复之类的”。另一方面，奥尔特曼表示马斯克“打击了我们一些关键研究人员的士气”，并对“组织文化造成了长期的巨大损害”。所以，一些人可能确实为摆脱了他而感到庆幸。

在整个庭审过程中，我看到马斯克一方的法律表现相当拙劣。在交叉质询中，我们经历了超过10分钟的折磨，史蒂文·莫洛向奥尔特曼列举了各种人称他为骗子：苏茨克维、米拉·穆拉蒂、海伦·托纳、塔莎·麦考利、丹妮拉和达里奥·阿莫迪（Anthropic的联合创始人，OpenAI前员工）、奥尔特曼第一家初创公司Loopt的员工、最近的《纽约客》文章、一本叫《乐观主义者》的书等等。莫洛确实通过询问奥尔特曼关于庭审证词的内容得了一些分——奥尔特曼说他没太关注。莫洛表现得好像这不可思议，难道没人向奥尔特曼通报过证词内容吗？这有点好笑，也有点让人厌烦。奥尔特曼保持了冷静，对于自己是否说谎这个焦点问题，他显得受伤而困惑。这也是交叉质询相对成功的一部分，之后注意力就迅速涣散了。我在这场庭审里见过不少马斯克那边糟糕的辩护，今天尤其差劲。有一次，当莫洛试图抓住奥尔特曼既是CEO又兼任公司董事这一点大做文章时，奥尔特曼如实说道——CEO几乎总是同时担任其所运营公司的董事。

这场审判的目的不在于赢——而在于惩罚奥尔特曼、布罗克曼和OpenAI。莫洛还提出了一个关于非营利组织募资的说服力不强的论点，即如果斯坦福大学每年能筹到30亿美元，OpenAI就应该保持非营利性质。暂且不谈斯坦福拥有成千上万毕业生的捐赠网络，它是一所学校，资本需求完全不同，且不与任何知名营利公司竞争。退一步说，就算某个募资天才接手OpenAI基金会：30亿美元也仅相当于微软最初两笔投资的总和，远不足以将OpenAI扩展至如今的规模。如果算力是构建AI模型的主要瓶颈，那么莫洛的论点本身就暗示了OpenAI作为纯粹的非营利组织永远无法成功——他其实是在为辩方说话。

但问题是，莫洛并不需要真的擅长这份工作，因为这场审判的目的不在于赢——尽管我相信马斯克不介意赢。关键在于惩罚奥尔特曼、布罗克曼和OpenAI。马斯克已经相当彻底地做到了这一点——在公众心中强化了奥尔特曼是骗子和蛇的形象。今天早上，我读到《华尔街日报》的一篇独家报道，称多名共和党州检察长和众议院监督委员会想要调查萨姆·奥尔特曼的投资。报道中充斥着对这场审判的引用。

是的，奥尔特曼在证人席上很有说服力。他甚至可能打赢这场官司。但看起来，马斯克的报复才刚刚开始。

中文翻译：

在听取了两周各色证人对他的评价——“他是个满口谎言的蛇”之后，陪审团终于听到了这条“蛇”本人的证词：萨姆·奥尔特曼。在证词陈述结束时，他的律师威廉·萨维特问他，被指控窃取一家慈善机构是什么感受。

萨姆·奥尔特曼在证人席上表现得很出色，但这或许还不够。
埃隆·马斯克可能已经对这位OpenAI首席执行官造成了更长期的声誉损害。

“我们通过大量的辛勤工作，创建了这个极其庞大的慈善机构，我同意，你不能就这样把它偷走。”奥尔特曼说，“但马斯克先生确实曾试图扼杀它，我猜，还不止一次。”奥尔特曼完全处于“来自圣路易斯的好孩子”模式，并勉强装出一副对自己遭遇感到困惑的样子。当他抱着一堆证据活页夹走下证人席时，他甚至看起来有点像个小学生。在直接陈述开始时，他似乎有些紧张，不过很快就放松了下来。总的来说，他的证词听起来是可信的——而且有时，陪审团似乎还挺喜欢他。

在整个审判过程中，我一直很难想象陪审团对这一切是如何理解的，因为我对这些出庭作证的人物过于熟悉了。我听到过一些在宣誓后说的离谱谎言，比如埃隆·马斯克向我们所有人宣称他从不发脾气（结果他在交叉质询时当场发飙）。又比如他几个孩子的母亲希冯·齐里斯告诉我们，她不知道马斯克正在创立xAI——这似乎与她自己的短信内容直接矛盾。或者当格雷格·“多少钱能让我赚到10亿美元？”·布罗克曼告诉我们他完全是为了使命时。我当然相信奥尔特曼不可信——我的意思是，《纽约客》曾刊登超过17000字的文章描述他有多么爱撒谎。但与马斯克不同的是，有同期文件可以支持奥尔特曼的说法。至少，大部分情况如此。

“我的看法是，他想要长期控制权”

在OpenAI的《Dota 2》获胜后，关于设立营利性分支的讨论正式开始。“马斯克先生非常强烈地认为，如果我们要成立一个营利性实体，他最初必须拥有完全控制权，”奥尔特曼说，“他只相信自己能做出那些最终会被证明正确的、非显而易见的决定。”

奥尔特曼作证说，他对马斯克坚持要求控制权感到不安，这不仅是因为马斯克不像其他人那样深度参与，更是因为OpenAI存在的意义就是不让任何一个人控制通用人工智能（AGI）。而在他曾担任总裁的创业孵化器Y Combinator，奥尔特曼见过太多关于控制权的争斗；当事情进展顺利时，没人愿意放弃权力。通过超级投票权之类的结构，创始人可以永远保持控制权。奇怪的是，奥尔特曼举的例子并非最知名的那个（如Meta的马克·扎克伯格），而是马斯克和他的SpaceX。当奥尔特曼向马斯克询问OpenAI的继任计划时，他得到了一个格外“令人毛骨悚然”的答复：马斯克说，如果他死了，“我没怎么想过这个问题，但也许控制权应该传给我的孩子们。”

我不太确定这一点。但我确实看到了一封2017年奥尔特曼写给齐里斯的邮件，其中写道：“我担心控制权问题。我认为任何人都不应该控制世界上第一个AGI——实际上，我们创办OpenAI的全部原因就是为了防止这种情况发生。”他接着表示，他并不介意眼前这个立即获得控制权的想法，并对“创造性结构”持开放态度——我理解这意思是，为了安抚马斯克，奥尔特曼愿意在公司发展到特定里程碑之前给予他控制权。

“我在里面读到了某种模糊的、轻微的威胁”

“我的看法是，他想要长期控制权，而且如果我们同意了他想要的那种结构，他本来是会得到这个权力的。”奥尔特曼在证人席上说。这听起来基本是对的。在后来播放的萨姆·特勒的证词录像中，我们听到马斯克不再投资任何他无法控制的项目。这也符合马斯克长期以来的执念——确保自己不会再像当年被赶出PayPal那样被赶出自己创立的公司。

马斯克还曾试图招募奥尔特曼加入特斯拉。我们看到了奥尔特曼和特勒之间的短信，特勒告诉奥尔特曼，马斯克决心无论如何都要加强特斯拉的AI实力，并希望奥尔特曼、布罗克曼和伊利亚·苏茨克维最终能加入。“我在里面读到了某种模糊的、轻微的威胁，意思是他要么带上你们，要么不带你们，都要在特斯拉内部搞这个项目。”奥尔特曼说。但他觉得特斯拉主要是一家汽车公司——允许它收购OpenAI将背叛OpenAI的使命。

后来，在特勒的证词中，我们看到了特勒在2018年2月4日凌晨12:40发给齐里斯的短信：“我不喜欢OpenAI在没有埃隆的情况下继续下去，”他写道，“我宁愿通过挖走领导层来搞垮它。”

当马斯克停止其季度捐赠时，OpenAI靠着“极其有限的现金周转期”和“微薄”的资金在运营。OpenAI确实有其他捐赠者，但没有人起诉它或加入马斯克的诉讼。（在法庭上未被提及的一份证据中的捐赠者是阿拉梅达研究公司，即山姆·班克曼-弗里德拥有的公司，此人目前因欺诈和洗钱罪在狱中服刑。）马斯克从董事会辞职意味着“人们怀疑他是否试图报复我们或其他什么。”另一方面，奥尔特曼表示，马斯克“打击了我们一些关键研究人员的积极性”，并“对组织文化造成了长期的巨大损害。”所以，似乎有些人确实因为摆脱了他而松了一口气。

在整个审判过程中，我看到马斯克一方的律师水平相当拙劣

我们看到了大量证据，表明在奥尔特曼设立OpenAI营利性分支的整个过程中，他都通过直接沟通、或经由齐里斯和特勒，及时向马斯克通报了进展情况。马斯克从未提出过异议，而且，无论他公开对微软的投资说过什么，有大量证据表明他在私下里是知情的。

在交叉质询环节，我们“有幸”观看了十多分钟，史蒂文·莫洛罗列了一长串指控奥尔特曼撒谎的人：苏茨克维、米拉·穆拉蒂、海伦·托纳、塔莎·麦考利、丹妮拉和达里奥·阿莫代（前OpenAI员工、Anthropic联合创始人）、奥尔特曼第一家初创公司Loopt的员工、最近那篇《纽约客》文章、一本名为《乐观主义者》的书，等等。莫洛通过询问奥尔特曼对庭审证词是否知情确实赢得了几分，奥尔特曼表示自己并未密切关注。莫洛表现得好像这难以置信：肯定会有人跟奥尔特曼通报证词内容的吧？

这既有点好笑，也有点让人厌烦。不过，奥尔特曼保持了冷静，似乎对大家纠结于他是不是个骗子感到受伤和困惑。这也是交叉质询中最成功的一部分，之后注意力便急剧下降。在整个审判过程中，我看到马斯克一方的律师水平相当拙劣，而今天尤其糟糕。有一次，当莫洛试图利用奥尔特曼同时担任CEO和公司董事的身份大作文章时，奥尔特曼如实说道——CEO几乎总是会兼任其所运营公司的董事。

（在我的笔记中，此时我写道：“天哪，莫洛真不擅长这个。”）

这场审判的目的不是为了赢——而是为了惩罚奥尔特曼、布罗克曼和OpenAI

还有一个关于非营利组织筹款的不具说服力的论点，即如果斯坦福大学每年能筹集30亿美元，那么OpenAI本应继续保持非营利性质。好吧，我们姑且想想。斯坦福大学拥有数千名毕业生组成的捐赠网络。它是一所学校，资金需求截然不同。它不与任何有信誉的营利性公司竞争。但撇开这些不谈，假设某个筹款天才接管了OpenAI基金会：30亿美元只是微软最初两笔投资的总和，根本不足以将OpenAI扩展到现在的规模。如果计算能力是构建AI模型的主要瓶颈，那么莫洛的论证思路实际上暗示，OpenAI如果仅靠非营利模式，永远不可能成功。他这是在替辩方说话。

但问题是，莫洛其实并不需要真正擅长这份工作，因为这场审判的目的不是为了赢——尽管我确信马斯克不介意赢。其目的是为了惩罚奥尔特曼、布罗克曼和OpenAI。马斯克已经相当彻底地做到了这一点——在公众心目中强化了奥尔特曼是骗子和小人的形象。今天早上，我在《华尔街日报》上读到一篇独家报道，称多位共和党州检察长和众议院监督委员会希望调查萨姆·奥尔特曼的投资情况。文章中充斥着对这场审判的引用。

所以，是的，奥尔特曼在证人席上很有说服力。他甚至可能赢得这场诉讼。但看起来，马斯克的报复才刚刚开始。

英文来源：

After two weeks of hearing from assorted witnesses that he was a lying snake, the jury finally heard from the lying snake himself: Sam Altman. At the end of the testimony, his lawyer William Savitt asked him how it felt to be accused of stealing a charity.
Sam Altman was winning on the stand, but it might not be enough
Elon Musk may have done more long-term reputational damage to the OpenAI CEO.
Sam Altman was winning on the stand, but it might not be enough
Elon Musk may have done more long-term reputational damage to the OpenAI CEO.
“We created, through a ton of hard work, this extremely large charity, and I agree you can’t steal it,” Altman said. “Mr. Musk did try to kill it, I guess. Twice.”
Altman was fully in “nice kid from St. Louis” mode, and did a passable impression of a man who was bewildered at what was happening to him. When he stepped down from the stand holding a stack of evidence binders, he even looked a little like a schoolboy. He seemed nervous at the beginning of his direct testimony, though he warmed up fairly quickly. Overall, he seemed to give credible testimony — and at times, it seemed like the jury liked him.
Throughout this trial I’ve had some difficulty imagining what the jury is making of all this because I am a little too familiar with the figures who are testifying. I have heard some audacious lies under oath, like when Elon Musk told us all he doesn’t lose his temper. (He then proceeded to lose his temper on cross-examination.) Or like when Shivon Zilis, the mother of several of his children, told us that she didn’t know Musk was starting xAI — which seemed to be directly contradicted by her text messages. Or when Greg “What will take me to $1B?” Brockman told us he was all about the mission. I certainly believe Altman isn’t trustworthy — I mean, The New Yorker published more than 17,000 words about how much he lies. But unlike with Musk, there are contemporaneous documents backing Altman’s version of the story. At least, mostly.
“My belief is he wanted to have long-term control”
After OpenAI’s Dota 2 win, discussions for a for-profit arm started in earnest. “Mr. Musk felt very strongly that if we were going to form a for-profit he needed to have total control over it initially,” Altman said. “He only trusted himself to make non-obvious decisions that were going to turn out to be correct.”
Altman testified that he was uncomfortable with Musk’s insistence on control, not just because Musk hadn’t been as involved as everyone else, but because OpenAI existed so no one person would control AGI. And at Y Combinator, the startup incubator where he was president, Altman had seen a lot of control fights; no one wanted to give up power when things were going well. With structures like supervoting shares, founders could retain control forever. Curiously, Altman’s example was not the most famous one (Mark Zuckerberg at Meta); it was Musk and SpaceX. When Altman asked Musk about succession plans for OpenAI, he got a particularly “hair-raising” answer: In the event of Musk’s death, Musk said, “I haven’t thought about it a ton, but maybe control should pass to my children.”
I don’t know about that. But I do know that I saw a 2017 email from Altman to Zilis in which he wrote, “I am worried about control. I don’t think any one person should have control of the world’s first AGI — in fact the whole reason we started OpenAI was so that wouldn’t happen.” He went on to say that he didn’t mind the idea of immediate control and was open to “creative structures” — which I understood to mean that, in order to placate Musk, Altman was willing to give him control up to specific milestones in company development.
“I read a vague, like, a lightweight threat in there”
“My belief is he wanted to have long-term control and that he would’ve had that had we agreed to the structure he wanted,” Altman said on the stand. This sounds basically right. In later video testimony from Sam Teller’s deposition, we heard that Musk no longer invests in anything he doesn’t control. This also fits with Musk’s long-term fixation on making sure he can’t get booted from his own company the way he got booted from PayPal.
Musk also tried to recruit Altman to Tesla. We saw texts between Altman and Teller, in which Teller told Altman that Musk was committed to beefing up Tesla’s AI no matter what, and that he hoped that Altman, Brockman, and Ilya Sutskever would want to join eventually. “I read a vague, like, a lightweight threat in there, that he’s gonna do this inside of Tesla with or without you,” Altman said. But he felt that Tesla was primarily a car company — allowing it to acquire OpenAI would betray OpenAI’s mission.
Later, in Teller’s testimony, we saw texts Teller sent to Zilis at 12:40AM on February 4th, 2018: “I don’t love OpenAI continuing without Elon,” he wrote. “Would rather disable it by recruiting the leaders.”
When Musk stopped his quarterly donations, OpenAI was operating on a “shoestring” with an “extremely short runway of cash.” OpenAI did have other donors, none of whom have sued it or joined Musk’s suit. (One donor in the exhibit that wasn’t called out to the courtroom was Alameda Research, the firm owned by Sam Bankman-Fried, who is now in prison for fraud and money laundering.) Musk’s resignation from the board meant “people wondered if he was gonna try to take, uh, vengeance out on us or something.” On the other hand, Altman said Musk had “demotivated some of our key researchers” and done “huge damage for a long time to the culture of the organization.” So it sure seems like some people were relieved to be rid of him.
I’ve seen some fairly shoddy lawyering from Musk’s side throughout this trial
We saw a lot of evidence that throughout the time Altman was setting up OpenAI’s for-profit arm, he kept Musk apprised of what was going on, either directly or through Zilis or Teller. At no point did Musk object, and whatever he said publicly about the Microsoft investments, there was plenty of evidence that privately he’d been made aware.
On the cross-examination, we were treated to more than 10 minutes of Steven Molo telling Altman that various and assorted people had called him a liar: Sutskever, Mira Murati, Helen Toner, Tasha McCauley, Daniela and Dario Amodei (former OpenAI employees and founders of Anthropic), employees at Altman’s first startup Loopt, that recent New Yorker article, a book called The Optimist, etc. Molo did score some points by asking Altman about testimony in the trial, which Altman said he wasn’t paying close attention to. Molo acted as though this was inconceivable. Surely someone had informed Altman of what was said?
It was a little funny and also a little tiresome. Altman kept his cool, though, seeming hurt and confused by the focus on whether he was a liar. It was also the most successful part of the cross, which declined in focus precipitously afterward. I’ve seen some fairly shoddy lawyering from Musk’s side throughout this trial, and today was pretty bad. At one point, when Molo was trying to capitalize on Altman being both CEO and on the company’s board, Altman said — truthfully — that CEOs are almost always on the boards of the companies they run.
(At this point in my notes, I had written, “Boy, Molo is not very good at this.”)
The point of this trial isn’t to win — it’s to punish Altman, Brockman, and OpenAI
There was also an unconvincing argument about fundraising in nonprofits, specifically that if Stanford could raise $3 billion a year, OpenAI should have remained a nonprofit. Okay, let’s just think about that for a minute. Stanford has a donor network of thousands of graduates. It’s a school, which has very different capital requirements. It is not competing with any reputable for-profit companies. But leave that all aside and assume that some fundraising genius took over at the OpenAI Foundation: $3 billion is the initial two Microsoft investments combined, and not enough to scale OpenAI to where it is now. If compute is the main bottleneck on building AI models, then Molo’s line of argument suggests OpenAI never would have managed to be successful as a nonprofit alone. He’s making the defense’s case for them.
But the thing is, Molo doesn’t actually have to be good at this job, because the point of this trial isn’t to win — though I’m sure Musk wouldn’t mind a win. The point is to punish Altman, Brockman, and OpenAI. Musk has done that pretty thoroughly — reinforcing in the public’s mind that Altman is a liar and a snake. This morning, I read an exclusive in The Wall Street Journal that assorted Republican AGs and the House Oversight committee wanted to look into Sam Altman’s investments. References to the trial are peppered throughout the article.
So yes, Altman was convincing on the stand. He may even win the suit. But it sure seems like Musk’s vengeance has just begun.