字号·· | 护眼
彭博社

Anthropic的达里奥·阿莫代伊并非总能如愿Anthropic’s Dario Amodei Can’t Always Have His Way

点「原文对照」整页切到原文,或双击某段只看那段的原文。

达里奥·阿莫代伊的专注既能助益,也会带来伤害。摄影师:大卫·保罗·莫里斯/彭博社Dario Amodei’s focus can help — and hurt Photographer: David Paul Morris/Bloomberg

PBC 受益于一位领导者的专注,这位领导者在决策上从不摇摆。但随着公司在市场动荡时期冲刺首次公开募股(IPO),这一优势可能转化为劣势:过去一个月纳斯达克指数下跌约 6%,英伟达首席执行官黄仁勋可能被罢免的传闻削弱了市场对人工智能板块的信心。

PBC has benefited from the single-minded focus of a leader who doesn’t waver from his decisions. But that advantage could become a liability as it speeds toward an initial public offering in the middle of a febrile market, with the Nasdaq having fallen about 6% in the past month and the possible ouster of Nvidia Corp.’s Jensen Huang sapping confidence in the artificial intelligence trade.

Anthropic 首席执行官达里奥·阿莫代伊被视为硅谷最执着、最有原则的科技高管之一。相比之下,山姆·奥特曼更为灵活,今年早些时候曾在五角大楼使用 OpenAI 技术的问题上妥协,而阿莫代伊则坚决反对。

Anthropic Chief Executive Officer Dario Amodei is seen as one of Silicon Valley’s most dogged and principled technology executives. Where the more flexible Sam Altman was willing to bend to the whims of the Pentagon earlier this year over its use of OpenAI’s technology, Amodei stood firm against it.

奥特曼探索过多条潜在营收路径——推进 ChatGPT 面向消费者的业务、与乔尼·艾维合作打造 AI 设备、投资机器人领域——阿莫代伊却让 Anthropic 激光般聚焦于向企业销售 AI 服务。

While Altman has explored several potential revenue paths — pursuing a consumer business for ChatGPT, building an artificial intelligence device with Jony Ive and investing in robotics — Amodei has kept Anthropic laser-focused on selling AI services to businesses.

他还在 Anthropic 维持着一种更严密的文化:员工须认同 AI 可能灭绝人类的信念,并认为公司必须竞相构建最强大的系统、以身作则树立安全标杆。前员工透露,这种意识形态上的高度一致虽使公司成为产品功能的高产创造者,但员工也感到无法质疑公司路线。

He also maintains a much tighter culture at Anthropic, one where employees are expected to adhere to the belief that AI could wipe out humanity and that the company must race to set an example on safety by building the most powerful systems possible. Former employees have told me that while being in ideological lockstep has made the company a prolific creator of product features, staff also feel they can’t question the company line.

奥特曼对 OpenAI 上市采取灵活态度,因财务和安全顾虑推迟了 IPO。阿莫代伊则坚持原计划,要在感恩节前、即 11 月 16 日那一周上市。考虑到科技股间歇性抛售、市场对 AI 行业能否证明其巨额支出合理性的疑虑日增,这一时间表显得格外引人注目。

Altman took a fluid approach to OpenAI’s IPO, delaying it amid financial and security concerns. Amodei, meanwhile, has stuck with his plan to go public before Thanksgiving, on the week of Nov. 16. That’s remarkable given the sporadic selling off of tech stocks and growing doubts about whether the AI industry can justify its enormous spending.

Anthropic 至少在价格上表现出了一定的灵活性。其预期估值约为 1.3 万亿美元左右,远低于几周前传出的 2 万亿美元数字。但其在时机问题上拒绝让步,可能对市场和经济产生重大影响,因为该公司如今已成为前沿 AI 业务以及为其提供动力所需的庞大基础设施的风向标。

Anthropic has at least shown some flexibility on price. Its expected valuation of about $1.3 trillion or so is well below the $2 trillion figure floated just weeks ago. But its refusal to budge on timing could have significant implications for the market and economy, with the company now a bellwether for the business of frontier AI as well as the vast infrastructure needed to power it.

不过,对 Amodei 妥协意愿的更大考验,可能要在 IPO 之后才会到来,届时他将不得不向公众投资者负责。Anthropic 的创始人将保留公司 50.1% 的投票权,且其大多数董事由一个为保护其安全使命而设立的特殊信托选出。该公司似乎想要华尔街的资金,却不愿让投资者在公司运营上拥有太多发言权。当然,考虑到硅谷惯于赋予创始人超常控制权 —— 看看 Meta Platforms Inc. 和埃隆·马斯克掌管的任何公司 —— 这并不罕见,但这也是 Amodei 坚定信念的表现。

The bigger test of Amodei’s willingness to compromise, however, may come after the IPO, when he has to answer to public investors. Anthropic’s founders are set to retain 50.1% of the company’s voting power and most of its directors are chosen by a special trust set up to protect its safety mission. The company seems to want Wall Street's money without giving investors much say in how it’s run. Of course, that’s not so unusual given Silicon Valley’s penchant for giving founders outsized control — see: Meta Platforms Inc. and any company run by Elon Musk — but it’s also a sign of Amodei’s conviction.

他长期宣称,Anthropic 正在引领一场关于安全标准的“向上竞赛”。比宿敌 OpenAI 更早上市,将有助于确立其作为引领世界走向更美好未来的 AI 公司的地位。

He has long said that Anthropic is leading a “race to the top” on safety standards. Going public before arch-rival OpenAI would help seal his company’s status as an AI firm leading the world toward a better future.

对这位 CEO 而言,这场竞赛关乎生存。他的信念源于他在攻读博士学位期间阅读了雷·库兹韦尔的著作《奇点临近》,并深度卷入了理性主义和有效利他主义运动,这些运动警告过 AI 末日。Amodei 宣扬,尽快构建超级智能系统是超越“最不谨慎行为体”的唯一途径。这种对 Anthropic 使命的道德确定感,使该公司在某些方面显得难以相处;几位开发者和企业顾问曾私下向我抱怨 Anthropic 的傲慢和僵化,例如其关于数据零日留存的政策。

For the CEO, the race is existential. His beliefs go back to when he read Ray Kurzweil's book, , as a PhD studentand becamedeeply involved in the rationalist and effective altruist movements that warned of an AI doomsday. Amodei preaches that building superintelligent systems as quickly as possible is the only way to get ahead of the “least cautious actors.”That sense of moral certainty about Anthropic’s mission has made the company somewhat difficult to deal with; several developers and corporate advisers have privately complained to me about Anthropic’s hubris and rigidity on matters like its zero-day retention policies on data.

对于阿莫代伊及其公司的财务状况而言,这种不妥协的态度在不断变化的 AI 格局中可能代价高昂。中期选举结果表明,建设数据中心将变得更慢、成本更高,而 Anthropic 已经在算力上的支出超过 OpenAI,因为它在基础设施和芯片上的投资起步较晚。事实上,其 S-1 注册文件显示,上一季度 Anthropic 的算力成本增速首次快于营收增速。

For Amodei and his company’s financial position, that inflexibility could be costly in a shifting AI landscape. The results of the midterm elections suggest that building data centers will now get slower and more expensive, and Anthropic is already paying more for compute than OpenAI, having invested in infrastructure and chips late. In fact, its S-1 filing showed that in the last quarter, Anthropic’s computing costs grew faster than its revenue for the first time.

尽管算力供应商勉强能满足当前需求,但这些巨额投资能多快实现变现,以及两三年后是否仍有需求,目前尚不清楚。

And while providers of computing power can barely keep up with current demand, it’s unclear how quickly these huge investments will be monetized and whether appetite will still be there in two or three years’ time.

Anthropic 之所以取得显著成功,是因为阿莫代伊早早认清并付诸行动,坚持其关于安全和企业级 AI 的信念。但屡次被证明是正确的也有负面作用:它可能滋生一种危险的信念,即坚持立场永远是答案。

Anthropic became remarkably successful because Amodei identified and acted early on his beliefs about safety and enterprise AI, and pursued them with conviction. But being repeatedly right has a downside: It can usher in the dangerous belief that holding your ground is always the answer.

虽然 Anthropic 愿意调整估值,表明其 CEO 在某些问题上愿意妥协,但控制权不一定在妥协之列。当他的原则与公众投资者的要求发生冲突时会怎样?对于这位极具个性的阿莫代伊来说,或许需要更大的灵活性。

While Anthropic’s willingness to adjust its valuation suggests its CEO is willing to compromise on certain things, control is not necessarily one of them. What happens when his principles clash with the demands of his public investors? Some greater flexibility may be in order for the very singular Amodei.

阅读媒体原文 ↗