Anthropic 提交 335 页 S-1 招股书，称 AI 或带来奇迹时代也可能构成生存风险

Investing in any stock is kind of like gambling in the sense that you’re betting your money on a company’s future performance. But how do you bet on a technology that could, in the worst-case scenario, “pose catastrophic or existential risks to humanity,” or best case, “bring forth an age of miracles”?

To some extent, Anthropic suggests in a letter at the top of its S-1 filing, you have to just believe.

Anthropic’s 335-page S-1 prospectus, filed Monday, is part of a steady hype escalation in Silicon Valley’s interactions with Wall Street. In some ways, it’s not even that bizarre; SpaceX IPO’d in June after filing an S-1 that referenced dinosaurs and “colonizing Mars.” But its sweeping language and sheer volume — including 94 pages of risk factors — is a sometimes overwhelming collision of AI industry rhetoric and a hard financial pitch. Out-of-this-world promises are integral to Anthropic’s survival as a company. That doesn’t make reading the S-1 any less surreal.

In the letter, Anthropic’s founders urge prospective stock owners “not to think of your purchase in purely transactional terms. We hope you will see it instead as an opportunity to be part of something greater.” The march toward advanced AI “is one of the seminal moments in human history,” the letter says. “Our aim is to help humankind chart a safe course through this technological evolution, open new frontiers of discovery, and bring forth an age of miracles.” But the S-1 warns that if things go wrong, the consequences could be disastrous — even an “existential” risk to humanity. Many dangers are laid out in exhaustive detail, including risks that models “behave in ways that no user or developer intended,” leading to “catastrophic” incidents that the entire insurance industry couldn’t afford to underwrite.

Tech startups preparing to go public haven’t always used the information overload approach. “In 1980, a highly touted company called Apple Computer went public. Its prospectus, which I have, is 47 pages long with zero risk factors,” said Jay R. Ritter, a professor at the University of Florida’s Warrington College of Business. “If you go back and read it, it’s incredibly simple.”

Ritter pointed to Facebook’s S-1 filing as a turning point. The company now known as Meta filed a 200-page prospectus with 22 pages of risk factors. CEO Mark Zuckerberg’s letter to investors was a manifesto on his move-fast-and-break-things approach to growth, which he dubbed “The Hacker Way.”

While the idea might be to cover all possible bases, “a problem with this bloat is that it’s become more and more difficult to find the nuggets of what’s important,” Ritter said. “You know, I don’t know if you’re like me, but the longer it is, the less inclined I am to read it.”

Anthropic’s founders explain in their investor letter that oversharing is part of their “anti-sell” strategy, designed to attract long-term investors who believe in their vision.

“It is very dangerous when people buy your company’s stock based on expectations you cannot fulfill or on incorrect predictions about how you will behave. This is in fact very similar to one of Anthropic’s guiding principles: underpromise and overdeliver,” the founders’ letter says.

Some of the statements in Anthropic’s S-1 read as surprising admissions. “We may make forward investments that may seem irresponsibly large,” for instance, or “model safety, interpretability, alignment, and welfare may lead us to limit or delay the deprecation of older models.” Maybe none of this should be surprising, considering the founders do warn readers about the company’s policy of practicing “radical honesty.”

Anthropic’s ‘radical honesty’ includes being open about the prospects of human extinction

This policy led to the biggest jumpscare of the whole filing, on page 46:

“Concerns have been raised both within our organization and by other researchers, regulators, policymakers, and industry participants that, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity.”

Followed quickly by:

“Our development of highly advanced models, platforms, and applications and expansion of use cases could further increase the risk that our models cause harm.”

Anthropic has famously been a proponent of doomsday AI scenarios. CEO Dario Amodei has said there’s an up to 25 percent chance that further AI development could lead to human extinction.

The heart of Anthropic’s pitch — or anti-sell — is that despite the risks, it is absolutely committed to developing safe and secure AI technology. After all, it will go public as a public benefit corporation, a type of business entity that, unlike C-corporations that are required to maximize shareholder value, must prioritize a social mission alongside shareholder interests.

“Powerful AI” — Anthropic’s term for hypothetical future AI that surpasses collective human intelligence — is presented in the company’s S-1 as inevitable. If this technology is coming no matter what, then Anthropic, a public benefit corporation, should lead the charge. But Anthropic will need a lot of money, possibly an unlimited amount, to do this. And more practically, as Liz Lopatto has written, powerful AI (in the form of dramatic recursive self-improvement) might be about the only way Anthropic can stay ahead of the competition and survive.

“All of us have a sincere conviction that the AI exponential will generate so much value, that, in the long run, it will far outweigh any shorter-term economic sacrifices we make as part of our mission,” the letter says.

For all the information the company made available in its S-1, there are some gaps. Crucial information that could be used to assess profitability — third quarter results, for instance — is missing. Anthropic saw explosive revenue growth in the second quarter, but the second quarter was also the height of tokenmaxxing, which could have led to a surge in sales. Some corporate executives have warned that AI costs are surging out of control. While Anthropic’s operating losses narrowed in the second quarter, the S-1 also revealed hundreds of billions of dollars in financial obligations to secure the computing power it says it needs to keep growing. It also said it will start building its own data center infrastructure, a costly move that would eat into profit.

Ritter thinks Anthropic will likely share its third quarter results the closer it gets to an IPO date. And besides, he added, “nobody knows” what will happen once a company goes public, including the company itself. It’s all a gamble in that sense.

Anthropic isn’t WeWork… although never making money is on its list of risk factors, of course

“For its first 22 years, Apple was largely a failure as far as investors were concerned,” Ritter said. “In its first three years, it greatly underperformed the market. And now it’s worth what, $5 trillion?”

When WeWork filed its S-1 in 2019, it was skewered for using lofty rhetoric. A line about the company’s aim to “elevate the world’s consciousness” was especially roasted.

That was different, Ritter said. WeWork was “a money-losing real estate company pretending to be a tech company at a ridiculously high valuation.”

“There was no chance of ever making huge profits in the future.”

Technically, that’s a risk for Anthropic too. It’s on page 63 of the S-1.

“We have a history of net losses and may not generate sufficient revenue to achieve or sustain profitability in the future.”

Then again, this is the age of miracles.

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

• Ellen Thomas

原文版权归 The Verge · AI 及原作者所有，本站转载仅供阅读。如需删除，请联系 support@aisa.one。