Pick your perdition - Anthropic's throwing dice like Pascal on his deathbed
Pick Your Perdition
Everyone's talking about the money. I'm stuck on page 80.
The coverage of Anthropic's leaked IPO prospectus has been almost entirely about capital. A $42 billion net loss (most of it on paper, an accounting artifact of investor obligations, but still). Nearly a quarter of its revenue coming from just two customers. A valuation pushing $2 trillion. Fine. Big numbers. But the line I can't stop rereading is this one, from CNN via Reuters:
Anthropic dedicated 80 pages to its risk factors in the leaked prospectus, significantly more than the 48 pages it spent detailing its business, according to Reuters. — CNN
Emphasis mine.
Eighty pages of ways this could go wrong. Forty-eight pages of what they actually do.
I'm not an AI doomer. I'm an anticapitalist, regulation-positive elder Millennial, which is arguably worse, because I think we'll find a way to end things regardless. Maybe it's AI. Maybe it's climate change. Maybe it's good old-fashioned global thermonuclear war.
Pick your perdition.
But that's the thing, isn't it? We pick it.
Pascal's Wager
Blaise Pascal's wager, published posthumously in his Pensées, is an early mash-up of probability, theology, and good old self-interest. It goes something like this:
| God exists | God doesn't exist | |
|---|---|---|
| Believe in God | Total salvation | Oblivion |
| Don't believe in God | Burnination | Oblivion (but make it smug) |
God either is or isn't. If God is and you live by His will, you're saved. If God is and you don't, total burnination. If God isn't, nothing you did matters much; you're headed for oblivion either way. So with infinite reward on one side, eternal torture on the other, and a shrug in the middle, the only rational bet is on God.
Here's the crucial part, and the part that matters for everything below: your bet doesn't change whether God exists. God is the fixed variable. You only choose how to respond to it.
p(doom)
"There's a 25% chance that things go really, really badly." — Dario Amodei, to Axios
He also put the odds of things going "really, really well" at 75%. A one-in-four chance of catastrophe, delivered with the same energy as a boring meteorologist. The modern wager goes like this:
invest in AI, because it might solve all of humanity's problems. You'd be a fool not to.
Draw it out and it looks like this:
| Invest in AI | Don't invest in AI | |
|---|---|---|
| AI won't destroy humanity | Make all the money (maybe, somehow; it hasn't yet...) | None of the money. FOMO sadness |
| AI will destroy humanity | Burnination | Burnination |
By this grid, investing is the obvious move: you lose nothing extra and might win big. Pascal would be proud.
The Bet Builds the God
That's the deception and the gamble
Pascal couldn't summon God by betting on Him. We can. We built this thing. Every dollar and every data center changes the odds of which row we end up in. The probability isn't fixed itself but the output of the gamble sure as hell is.
So the table doesn't have two columns. It has at least three:
| Charge ahead | Pace it, regulate it | Don't build it | |
|---|---|---|---|
| AI won't destroy humanity | All the money (maybe) | Most of the money, a bit later | None of the money |
| AI will destroy humanity | Burnination | Burnination | Burnination |
| Odds of the bottom row | Highest | Lower | Lowest |
That last row is the one the "you'd be a fool not to invest" crowd leaves off. It's the whole game. Framing this as all-AI-or-no-AI is a false choice, and a convenient one.
The Prospectus Is the Bottom Row Signed, Stamped, and Notarized
Pascal never had to get his wager past the SEC but Anthropic did. The company wrote the bottom row of the table down, in a legal filing, under its own name, and is asking you to buy in anyway. That's what gets me about those 80 pages.
To be fair: a risk-factor section is a legal shield. Companies list every plausible disaster so they can't be sued later. Disclosing a risk isn't predicting it, necessarily. But it's not nothing either. Someone at that company decided "catastrophic or existential risk to humanity" was material enough to tell investors about.
And to be fair again: Dario Amodei published an essay three weeks ago literally titled "We Must Pace the Frontier," calling on the whole industry to slow down. I agree but I don't believe it is what he, or the industry, want. I think they're trying to quell you and me, maybe regulators.
None of the pacing call stops training or shipping, conveniently. It doesn't stop a $2 trillion IPO. The pacing essay is the hedge: the middle column, carefully chosen so the money never has to sit out. Everyone gets to be on record warning about the bottom row while still placing the bet that makes it more likely.
It's just capitalism doing what capitalism does. Chose the option where the wealth keeps flowing but call it prudence.
Anthropic, OpenAI, all of them: they're not cartoon villains. They're something more ordinary and more dumbly dangerous. They're so thoroughly trained by the market that "don't extract wealth from this" isn't a column they can see. Even when the downside is the one they wrote down themselves. And one that we're bringing on ourselves.
Pick your perdition. Just don't pretend you didn't pick it.