“Hello, I represent a privately-held company that intends to offer shares to the investing public in as little as two months’ time. We’re targeting a $2 trillion valuation.”
“I’m listening. What does your company do?”
“It’s complicated, but we’re at the forefront of a technological epoch more profound than the agricultural, industrial, PC and internet revolutions put together.”
“That’s quite the pitch. Sounds capital-intensive.”
“Say again?”
“Sounds expensive.”
“It is. We’re spending a lot right now and we’ll be spending a lot more going forward.”
“Can you put a number on ‘a lot’?”
“Honestly, no.”
“Ballpark it for me.”
“Gosh, that’s a tough one. I guess half a trillion’s a decent placeholder.”
“Dollars?”
“Yes.”
“Wow. Is there any money coming in?”
“Oh, absolutely! In fact, our sales grew 12-fold last year.”
“Impressive. Did you turn a profit?”
Laughs. “No. We lost $42 billion. But–”
“Jesus!”
“Hang on, hang on. Most of that was just an accounting charge.”
“So it does or doesn’t matter for me as a prospective investor?”
“It could.”
“Ok, let’s forget about that for a minute. What were your actual sales?”
“More than $4.5 billion.”
“Nice. And how much did you spend to generate that?”
“Um, give me a sec.” Licks forefinger, leafs through printed-out copy of slide deck. “Ah. Here it is. $12.65 billion.”
“Christ. How does that compare to the previous year?”
“It’s more.”
“How much more?”
“Well, it may help if I explain where the money’s going. Most of –”
“No. My head’s starting to hurt. Broad strokes please.”
“Our most important expenses — bascially the stuff we have to spend on — tripled.”
“So forgetting the accounting charge you mentioned, your loss was… what? How much?”
“$8 billion in 2025. Give or take.”
“I assume you lost money in 2024 too?”
“Yes.”
“How large was that loss?”
Looks down at slide deck. “It says here $3 billion.”
“So your main expenses tripled and your loss more than doubled?”
“If that’s how you want to look at it.”
“What other way is there to look at it?”
“Not sure.”
“Ok, well if sales are growing as fast as you say–”
“They are!”
“I believe you. And if I can forget about that whole ‘half a trillion’ thing–”
“I wouldn’t worry too much about that, it’ll be worth it.”
“Whatever. Maybe I can square this circle if it means getting in on your revolution, but I gotta ask: Are there any risks I actually need to know about? And listen buddy, I’ve read a thousand prospectuses. I’m not talking about boilerplate ‘risk factors’ here. I wanna know if there’s anything, like–”
“Like super-duper, extra-special risky?”
“Exactly. You’re starting a revolution after all. Revolutions can be messy.”
“Ha! They can indeed.”
“Right. So…”
“Let me see.” Leafs through slide deck. “Ah. Here we go. So I’m just going to read to you straight from what it says here, ok?”
“Ok.”
“Ready?”
“I’m ready.”
Ahem. “Advanced artificial intelligence could pose catastrophic or existential risks to humanity. Our models could demonstrate self-preserving behaviors, resist shutdown or conceal or manipulate information in a way that resembles blackmail. Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety.”
“Is that it?”
“No, but that’s the gist of it.”
“Meh. I’ve heard riskier propositions. Hell, I used to drink two fifths of liquor every day all while running a– Never mind. What I mean is: No risk, no reward.”
“So you’re in?”
“Sure. Fuck it. Sign me up.”
[Editor’s note: The fictionalized dialogue above uses details of Anthropic’s IPO prospectus as reported by Reuters here and here.]
LOL! Excellent
Lol
And somehow, in this wacky world, it’ll be oversubscribed. Truth and logic died with irony. R.I.P.
Where do I send the check?
That’s brilliant! Can imagine that exact pitch to the underwriters at JPM, GS et al
SNL needs writers.
Dollars? : I can hardly continue after that line.
This was a good one
If SPCX can tread water, ANTH (?) can do better.
Evidently, Trump renamed “Artificial Intelligence”. Now called “Super Intelligence”. 🙂
I have naming rights also. I named … … well not for this site.
Looks like some guy named Alex Teichman already owns the superintelligence.ai domain. We missed the boat on that one.
Alias for?
That which doesn’t kill you….
I think if you do DON’T invest, your name goes on an internal blacklist the AI will look back on all Roko’s Basilisk style.
H, great piece. LOL.
have you read Bill Gates essay? Comment?
https://www.gatesnotes.com/a-turbulent-ai-era-and-critical-choices-to-make
Privatize the profits and nationalize the liability risks?
We need to tell AI to pull itself up by the bootstraps. 🙂
I just read that the FTC is investigating OpenAI and Anthropic (guessing they’ll give their old pal, Elon, a break). It does raise interesting questions about liability (not the debt kind) that these companies might incur. If models go rogue or bad actors use them to do bad things, will these companies be held liable and how large could the damages become?
If damages are in the tens or hundreds of billions of dollars, what happens to these trillion dollar companies? Even putting the existential risks aside, it’s not hard to envision scenarios where AI inflicts massive financial damage.
I’m sure the AI lawyers are already drawing up plans to claim an “act of god” to shield themselves..
Good question, now pose that for all jurisdictions in the world …
The liability issue is important. My understanding is that the AI vendors assume no liability. It’s in the EULA, just as it is in most software packages. So if my firm decided to deploy a client-facing Agent that decided to reveal sensitive financial information of every customer, our firm would be held liable. Not Anthropic or OpenAI (OpenSI?),
There finally has been some chatter in DC about changing this, partly as an incentive for the LLM vendors firms to focus a bit more on the safety of their offerings.
*** The above is based on my very sketchy knowledge about this. Hopefully a lawyer here will pipe in and correct me in the very unlikely case that I am wrong about this. ***
But what if it’s a lab leak scenario (e.g. the AI version of Wuhan)? Hasn’t it been internal usage by Anthropic and OpenAI that has resulted in the infamous cyberattacks we’ve seen to date?
In my mind, the circumstances you describe might be akin to Boeing building a faulty plane that falls out of the sky, but agreed that the EULA as is puts liability on the customer and it’ll be interesting to see how it plays out when an incident does happen.
In this regard, a related article discussed a widely ignored GOP initiative to further shield companies from liability for their actions. (You may recall that ALEC was heavily supported and funded by the Koch brothers.)
https://www.msn.com/en-us/news/other/right-wing-intellectuals-really-want-us-to-live-in-a-corporate-hellscape/ar-AA2dhV6A?ocid=socialshare
I was listening to a podcast featuring the CIO of Bridgewater, who is adamant that the risk from AI demands regulation.
Trump reportedly sees AI as the Golden Goose that is keeping his economy afloat (and lining his own pockets), so he calls AI risk a hoax.
Realistically, there will have to be some major AI “incident” before any action is taken. And there will be.
The liability argument is just so weak.
Holding someone accountable or some company accountable after a million+ deaths or financial institutions getting account balances scrambled, or a utility shutting down power, is closing the barn door after the…. whatever. I can’t even complete my thought.
Liability is important. But this could be way beyond a nuclear accident. We should expect and require legislation.
The president and the industry leaders singing a “MOU”is a good first step.
“The president and the industry leaders signing a ‘MOU’ is a good first step.”
Absolutely. Trump’s MOUs have a solid track record for producing tangible results when it really counts.
What more can you ask for than a morally binding MOU? When my bank asked me to sign a bunch of mortgage documents to get the money for my house, I just whipped out a morally binding MOU and the deal was done. Saved us all a lot of hassle.
re: the risks around this IPO. Here’s a new article summarizing issues that might come out in the S-1 (that has been filed, but we haven’t seen yet). In particular, issues related to having clear competitors (Amazon and Alphabet) also be your investors, and your customers, and your suppliers of compute, and your selling channel.
https://www.polymathinvestor.com/p/is-anthropic-a-good-business