OpenAI的总裁“什么都做”，唯独不回答问题。

qimuai 发布于 阅读：0 一手编译

OpenAI的总裁“什么都做”，唯独不回答问题。

内容来源：https://www.theverge.com/ai-artificial-intelligence/923684/musk-brockman-altman-openai-trial

内容总结：

法庭直击：OpenAI总裁格雷格·布罗克曼出庭作证，马斯克律师团直指其“贪婪”与“不可信”

在埃隆·马斯克起诉OpenAI一案的庭审中，OpenAI总裁格雷格·布罗克曼成为迄今为止最关键的证人——但这位证人的表现，却让旁观者直呼“不可信赖”。

法庭上的“辩论冠军”：连冠词都不放过

布罗克曼的作证方式极为罕见：他被首先进行交叉质询，随后才是直接质询。整个过程中，他展现出浓郁的高中辩论队“较真”风格，频繁使用“我不会那样描述”、“我不会那么说”、“这听起来像我写的，能看看上下文吗？”等措辞。当马斯克的律师史蒂文·莫洛宣读证据时，布罗克曼会学究式地纠正律师漏掉的每一个词，哪怕只是“a”或“the”这样的冠词。

在被问及微软100亿美元投资是否是OpenAI最大的财务事件时，布罗克曼的回答是：“这是唯一一笔100亿美元的投资。”这种“文字游戏”式的回应让现场气氛颇为尴尬。

日记曝光：2017年已现“逐利”之心

比布罗克曼的法庭表现更致命的，是他电脑中的一系列日记文本。这些2017年前后的记录，清晰暴露了他早期对非营利使命的“机会主义”态度。其中一条写道：“顺便说一句，另一个感悟是，从他（指马斯克）那里窃取非营利成果是错的。在他不知情的情况下转变为公益公司，那将是极其道德破产的行为，而他真的不傻。”另一条则直言：“也许我们应该直接转为营利公司。为我们自己赚钱听起来很棒。”还有一条写道：“不能说我们致力于非营利。不想说我们承诺了。如果三个月后我们搞了公益公司，那就是谎言。”

马斯克律师团队称，这些日记直接印证了马斯克“窃取慈善成果”的指控。

股权之问：30亿美元身家为何不捐？

庭审中，莫洛律师的一个问题让布罗克曼陷入困境。已知布罗克曼在OpenAI营利性部门的持股价值约300亿美元，而他在日记中曾写下“什么能让我赚到10亿美元？”的疑问。莫洛追问：“既然10亿美元对你来说就足够了，为什么没有将290亿美元捐给OpenAI的非营利部门？”布罗克曼并未正面回答，而是用关于非营利部门在营利部门中持股价值的“废话”搪塞。经过数轮“网球赛”式的拉锯，布罗克曼始终未能给出答案。

“紫色盒子”之争：细节拉锯令人无奈

另一个令人费解的细节是，莫洛问及OpenAI是否常用“紫色盒子”来标注重要事项，布罗克曼回答“不是”。但随后出示的文件显示，OpenAI确实在员工和投资文件中使用紫色盒子来突出重要内容。记者在笔记中无奈写道：“我们为什么要为那该死的紫色盒子争来争去？”

多重利益关联：与奥特曼的财务纽带

布罗克曼还被揭露与OpenAI多家供应商存在财务利益关联，包括Cerebras、CoreWeave、Stripe和Helion Energy。更有甚者，他与OpenAI CEO萨姆·奥特曼存在直接财务联系：布罗克曼持有奥特曼家族办公室1%的股份，这是作为他放弃Y Combinator股票的补偿——因为“Y Combinator的股票已经用来满足其他员工的入职条件了”。2017年的一封邮件显示，马斯克对此并不知情，他转发马斯克助理的邮件给布罗克曼，连打了两个问号。

“我做所有事”——总裁的自辩与叙事

在随后的直接质询中，布罗克曼被问及作为总裁的职责时，回答：“我做所有事。”这一回应让旁听者直呼“崩溃”。他开始讲述OpenAI的创立故事：他和奥特曼是发起人，两人在离开Stripe时萌生想法，通过一系列温馨晚餐、纳帕谷之旅、AI会议等场景逐步推进。然而，在布罗克曼展示的OpenAI第一天工作照中，有奥特曼，却没有最重要的投资者马斯克。

马斯克的形象：遥远而“执念”于对手

在布罗克曼的叙述中，马斯克被描绘成一个遥远甚至有些可怖的形象。在一次晚餐中，马斯克直接问谷歌的丹尼斯·哈萨比斯是否“邪恶”。事实上，马斯克似乎对哈萨比斯“非常执著”，却几乎从未提及拉里·佩奇——而佩奇在马斯克口中正是OpenAI成立的导火索。OpenAI联合创始人伊利亚·苏茨克维曾给布罗克曼发短信说：“马斯克可能每周只来半天。但我担心我们的工作环境会变得非常紧张。”

结语：陪审团面临艰难抉择

庭审在马斯克律师的猛烈攻势中暂告一段落。从布罗克曼的日记内容、法庭上的“较真”态度，到面对关键问题时的回避，再到与奥特曼及其关联公司的财务纽带，马斯克团队成功塑造了一个“精明且不可信”的证人形象。而布罗克曼随后讲述的“理想化创业故事”，又与日记中的逐利心态形成鲜明对比。

正如观察者所言：“陪审团必须在两个都不太值得信任的男人之间，决定谁更值得信任。这个任务，我可不想替他们做。”

中文翻译：

迄今为止，埃隆·马斯克起诉OpenAI一案中最有力的证据，当属格雷格·布罗克曼的日记。而布罗克曼本人，则紧随其后，堪称第二号“有力证据”。
OpenAI总裁“凡事都做”，唯独不回答问题。
对格雷格·布罗克曼来说，任何细节都值得争论，无论多么微不足道。
OpenAI总裁“凡事都做”，唯独不回答问题。
对格雷格·布罗克曼来说，任何细节都值得争论，无论多么微不足道。
布罗克曼的出庭方式相当不寻常——他先接受交叉质询，再进行直接质证——而且他身上有股浓厚的高中辩论社气息。他频繁使用“我不会那样描述”“我不会那样说”以及“那听起来像是我写的。我能看看上下文吗？”当马斯克的律师史蒂文·莫洛大声朗读某些证据时，如果莫洛跳过了一个单词，哪怕是像“一个”或“这个”这样的词，布罗克曼也会迂腐地纠正他。当被问及微软的100亿美元投资是否是OpenAI最大的财务事件时，布罗克曼回答说那是唯一一笔100亿美元的投资。拜托。
我之前说过，如果你能定义“认识论”这个词，你就不应该为自己出庭辩护。所以律师只是跳过了一个词——真的值得占用陪审团的时间来告诉我们所有人这个吗？把“全世界最聪明男孩”的名号留给你父母吧。
“那在道德上就太破产了。”
本来这已经够糟了。但这些日记条目——来自他电脑的一系列文本文件——更糟糕，因为它们清晰地揭示了布罗克曼至少从2017年左右开始的贪婪和机会主义。请看这一条：“顺便说一句，由此得出的另一个认识是，从他那里窃取非营利组织是错误的。在没有他的情况下将其转变为公益公司。那在道德上就太破产了，而且他真不是个傻瓜。”再看这一条：“也许我们应该直接转为营利性公司。为我们赚钱听起来很棒。”还有这条：“不能说我们致力于非营利组织。不想说我们致力于此。如果三个月后我们成立了公益公司，那就是撒谎。”
我注意到，“从他那里窃取非营利组织是错误的”这句话，与马斯克声称的“窃取慈善机构”的说法非常接近。
我们还没有完成直接质证，所以我确信我们将会听到一些关于激发这些条目的相关事件的开脱之词。但基于布罗克曼在交叉质询中的态度以及那些日记内容，我想我不会在去洗手间时放心让他看管我的包。
马斯克的团队正试图将布罗克曼描绘成贪婪之人，这点我信服。布罗克曼日记中那句臭名昭著的“怎样才能让我赚到10亿美元？”也出现了。我们确定，布罗克曼在OpenAI营利部门所持股份的价值约为300亿美元。莫洛问布罗克曼，既然10亿美元对他来说就够了，那他为什么不把290亿美元捐给OpenAI的非营利部门？
“我们到底为什么要为那个该死的紫色框框吵架？”
布罗克曼本可以这样说：“如果我一次性抛售所有持股，OpenAI实际收到的将远低于390亿美元，因为这是供求法则。”他或许会说：“让我自己也承担风险，对其他投资者来说是一个重要的信号。”或者，“那只是我账面上的净资产，不是真实的。”
他什么都没说。布罗克曼用一堆关于非营利部门在营利部门中持股价值多少的废话来回应。莫洛说这没有回答他的问题，并再次追问。我们就此来回拉扯了相当长一段时间；陪审员们的头来回转动，仿佛在看一场网球比赛。布罗克曼始终没有回答那个问题。
任何细节都值得争论，无论多么微不足道。莫洛问紫色框框是否是OpenAI通常用来引起对重要事项注意的手段，布罗克曼说不是。然后我们所有人都从文件中读到，OpenAI通常在员工和投资者文件中使用它们来突出重要事项。在我的笔记中，我写道：“我们到底为什么要为那个该死的紫色框框吵架？”
莫洛还通过提出OpenAI与布罗克曼持股的多家公司（Cerebras、CoreWeave、Stripe 和 Helion Energy）达成的各种交易，打出了另一记重拳。考虑到使用Stripe的公司数量之多，其与OpenAI的交易似乎微不足道——但OpenAI的承诺对Cerebras和CoreWeave两家公司都至关重要。
“我什么事都做。”
布罗克曼与奥特曼还有着直接的经济联系，这源于他们创办OpenAI时奥特曼提供给布罗克曼的一揽子薪酬方案。他持有奥特曼家族理财室1%的股份，这是布罗克曼作为放弃Y Combinator股票的替代方案获得的，因为“我们用来满足其他员工入职条件的Y Combinator股票已经用完了”。在一封2017年的邮件中，马斯克的得力助手贾里德·伯查尔写信给马斯克，称奥特曼向他透露了此事。马斯克将这封伯查尔的邮件转发给布罗克曼，并附上了“？？”。显然马斯克此前不知道这笔交易，布罗克曼不得不解释一番。
我之所以详尽叙述布罗克曼让自己变得不可信任的各种方式，是因为观看有权势的男人坐立不安很有趣。但这影响了我对他随后开始的直接证词的观点。布罗克曼首先讲述了OpenAI的创立故事，听起来像是经过上千场播客和主题演讲打磨过的。当被问及作为OpenAI总裁他做什么时，他回答说：“我什么事都做。”要不是在法庭上，我早就尖叫了。千禧一代的词汇表真是个悲剧。
按照这个说法，OpenAI是布罗克曼和山姆·奥特曼的主意。布罗克曼在离开Stripe时曾告诉奥特曼他对AI的兴趣（“我正在考虑搞一个人工智能项目，”奥特曼显然回答说，“我也正在考虑搞一个人工智能项目。”）。他们保持联系。最初的想法是建立一个Y Combinator的研究部门，但马斯克否决了，因为他不想与Y Combinator有关联。
马斯克对哈萨比斯似乎“非常执着且念念不忘”。
想象一下温馨晚餐、纳帕谷之旅（“我们的货车堵在路上一个半小时，但没人注意到”，因为谈话太精彩了）、AI会议等画面的蒙太奇。哇！真是太棒了！大家相处得如此融洽，创造力如此充沛！我们被迫听了一段漫长的关于伊利亚·苏茨克维犹豫是否要离开谷歌的叙述，接着是布罗克曼拍摄的OpenAI创立第一天的照片，所有人都在他的公寓里工作。（照片中有：奥特曼。缺席：马斯克。）我想你明白我的意思；我当然也明白了布罗克曼的意思。这是奥特曼和布罗克曼的心血。直到马斯克与布罗克曼和奥特曼召集的团队进行了最后的通话之后，马斯克才告诉他们，他想更多地参与进来。
在证词中，马斯克表现为一个疏远且时而令人畏惧的人物。在一次晚餐中，他问谷歌的德米斯·哈萨比斯是否是邪恶的。事实上，马斯克对哈萨比斯似乎“非常执着且念念不忘”，却从未提及拉里·佩奇，而按照马斯克的说法，拉里·佩奇才是OpenAI得以成立的原因。在苏茨克维发给布罗克曼的短信中，苏茨克维写道：“埃隆可能每周会和我们待半天。我想象了一下那会是什么情景，我担心我们的工作环境会变得压力山大。”
苏茨克维的担忧是对的；马斯克是出了名的难相处。我想明天我们会听到更多相关内容。但就目前情况来看，陪审团将不得不在两个都不太值得信任的人之间，决定更信任哪一个。我不同情他们这项任务。

最受欢迎

英文来源：

The strongest witness for Elon Musk’s case against OpenAI so far has been Greg Brockman’s journal. Brockman himself is running as a close second.
OpenAI’s president does ‘all the things,’ except answer a question
No detail was too small to argue over for Greg Brockman.
OpenAI’s president does ‘all the things,’ except answer a question
No detail was too small to argue over for Greg Brockman.
Brockman was called to the stand in a rather unusual way — he was cross-examined first, followed by a direct examination — and he had some serious high school debate club energy. There was a lot of “I wouldn’t characterize it that way,” “I wouldn’t say it that way,” and “That sounds like something I wrote. Can I see it in context?” When Musk’s attorney, Steven Molo, read some of the evidence aloud, Brockman would pedantically correct him if he skipped a word, even if that word was “a” or “the.” When asked if Microsoft’s $10 billion investment was the biggest financial event at OpenAI, Brockman replied it was the only $10 billion investment. Come on.
I have previously said that if you can define the word “epistemology,” you should not testify in your own defense. So the lawyer skipped a word — is it really worth taking up the jury’s time to tell us all that? Save being the world’s cleverest boy for your parents.
“that’d be pretty morally bankrupt.”
This would have been bad enough. But the journal entries — a series of text files from his computer — were worse, because they were very clear about Brockman’s greed and opportunism at least circa 2017. Here’s one: “btw another realization from this is that it’d be wrong to steal the non-profit from him. to convert to a b-corp without him. that’d be pretty morally bankrupt and he’s really not an idiot.” Here’s another: “maybe we should just flip to a for-profit. making money for us sounds great and all.” There’s also this: “cannot say we are committed to the non-profit. don’t wanna say we’re committed. if three months later we’re doing a b-corp it is a lie.”
“It’d be wrong to steal the non-profit from him” is very close to Musk’s “steal a charity” line, I notice.
We haven’t finished the direct examination yet, so I’m sure we’ll be hearing something exculpatory about the events that inspired these entries. But between Brockman’s attitude toward the cross and the journal entries, I don’t think I’d trust him to watch my bag while I used the restroom.
Musk’s team is trying to paint Brockman as being greedy, which I buy. The infamous “What will take me to $1B?” from Brockman’s journal made an appearance. We established that Brockman’s stake in OpenAI’s for-profit was worth about $30 billion. Molo asked Brockman why he hadn’t donated $29 billion to OpenAI’s non-profit arm if $1 billion was enough for him.
“Why are we fighting about the fucking purple box?”
Brockman could have said something like, “If I dumped all my holdings all at once, OpenAI would receive a lot less than $39 billion, because that’s how supply and demand works.” He might have said something like, “It’s an important signal to other investors for me to have skin in the game.” Or maybe, “That’s just my net worth on paper. It’s not real.”
He didn’t do any of this. Brockman replied with nonsense about how much the non-profit’s stake in the for-profit was worth. Molo said that didn’t answer his question and asked again. We went back and forth on this for quite some time; the jury’s heads snapped to and fro as though they were watching a tennis match. Brockman never did answer the question.
No detail was too small to argue over. Molo asked if purple boxes were something OpenAI generally used to call attention to something important, and Brockman said no. Then we all read in the document that OpenAI generally used them in employee and investor paperwork to highlight important things. In my notes I have written, “Why are we fighting about the fucking purple box?”
Molo landed another major blow by bringing up the various deals that OpenAI had with companies Brockman had a stake in: Cerebras, CoreWeave, Stripe, and Helion Energy. Given the sheer number of companies that use Stripe, its OpenAI deal seems piddling — but the OpenAI commitments seriously matter to both Cerebras and CoreWeave.
“I do all the things.”
Brockman also has direct financial ties to Altman because of a compensation package he was offered when they started OpenAI. He holds a 1 percent stake in Altman’s family office, which Brockman got in lieu of Y Combinator stock because “we ran out of Y Combinator stock fulfilling other [employees’] offers.” In a 2017 email, Musk’s bodyman, Jared Birchall, writes to Musk that Altman disclosed that to him, and Musk forwards the Birchall email to Brockman with a “??” Apparently Musk didn’t know about the deal, and Brockman had to explain it.
I am dwelling on the various ways that Brockman made himself untrustworthy because it’s fun to watch powerful men squirm. But it also has colored my view of his direct testimony, which started afterwards. Brockman began by telling a story of OpenAI’s founding that sounded like it had been polished for a thousand podcasts and keynote speeches. When asked what he did as the president of OpenAI, he replied, “I do all the things.” If we had not been in a courtroom, I would have screamed. Millennial vocabulary is a fucking tragedy.
In this telling, OpenAI was Brockman and Sam Altman’s idea. Brockman had told Altman about his interest in AI as he was leaving Stripe (“I’m thinking about doing an AI thing,” to which Altman apparently said, “I’m also thinking about doing an AI thing.”). They kept in touch. The original idea was purportedly to have a Y Combinator research arm, which Musk shot down because he didn’t want to be affiliated with Y Combinator.
Musk seemed “very consistent and fixated” on Hassabis
Imagine a montage of cozy dinners, trips to Napa (“our van got stalled for an hour and a half in traffic and no one noticed” because the conversation was so good), AI conferences. Gee whiz! It was so neat-o! Everyone got along so well and had such great creative energy! We were treated to a very long retelling of Ilya Sutskever waffling about leaving Google, and then a photo Brockman took of the first day of OpenAI, with everyone working from his apartment. (In the photo: Altman. Missing: Musk.) I think you get my drift; I certainly got Brockman’s. This was Altman’s and Brockman’s baby. It was only after Musk had done closing calls with the team that Altman and Brockman had assembled that Musk told them he wanted to be more involved.
Musk appeared in the testimony as a distant and at times menacing figure. At one dinner, he asked if Google’s Denis Hassabis was evil. In fact, Musk seemed “very consistent and fixated” on Hassabis, and never so much as mentioned Larry Page, who in Musk’s telling was the reason OpenAI came to be. In text messages from Sutskever to Brockman, Sutskever wrote, “Elon might spend half a day a week with us. I imagined how it will be and I worry that our work environment can become very stressful.”
Sutskever was correct to worry; Musk is famously difficult. I imagine we will hear more about that tomorrow. But as it stands so far, the jury will have to decide who of two not-especially-trustworthy men it trusts more. I don’t envy them the task.
Most Popular

ThevergeAI大爆炸

文章目录

    📚 推荐阅读


    扫描二维码，在手机上阅读
    版权所有：启幕AI
    文章标题：OpenAI的总裁“什么都做”，唯独不回答问题。
    文章链接：https://news.qimuai.cn/?post=3990
    本站文章均为原创，未经授权请勿用于任何商业用途