site banner

Scott Alexander on Sam Bankman-Fried, FTX and Effective Altruism

astralcodexten.substack.com

I made this a top level post because I think people here might want to discuss it but you can remove it if it doesn't meet your standards.

Edit: removed my opinion of Scott from the body

15
Jump in the discussion.

No email address required.

...

#7 is worrying. Yes, it's very hard to mathematically discover deep fraud that all of the experts didn't. Yes, you're not going to go to jail over failing to discover this stuff, and to the extent people are trying to pay it back that's more out of ethics than legal or even social obligation. And, yes, you're going to have worse tools than the people who's job is to work on this all the time. But to borrow from Sonya, sometimes you have to think about this stuff, with your own brain, in your own head.

(And as much as you don't have as many tools or expertise as ProPublica, you also aren't facing as high a standard at noticing a problem, and your opponent isn't going to put as many resources into snookering you.)

And it's not like "going to jail" is the only failure mode! Even if the 'only' problem with FTX was a hair-brained scheme that was simply going to go bankrupt without any fraud, it kinda is a problem for the charities and business working with them. At minimum, there's tremendous loss of cash and face, and the wide variety of problems of busted financials people are facing right now, but there's also potential problems when you try to actually cash a check that doesn't exist for services you've already rendered. That's not limited to speculative finance giving to charity, or even to private business working with other private business: I've seen companies completely fucked over because a business or even government decided that they weren't going to pay a big and important bill after the company had already put thrown together a lot of purchases or even fully delivered a service. In the extremes, a very fraudy company can end up with their partner's self-funded and pre-existing assets confiscated by creditors, and even if the paperwork still has the original owners in the 'right', it's still years of legal fees and missing (and sometimes damaged) assets. (In really severe extremes, you find out that the sponsor you were looking at has ties to stationary bandits of the more immediate sense, or that starts to trigger FCPA.)

You absolutely need to consider this stuff.

It's possible (maybe likely) that people who did that sort of serious consideration would still have have shrugged and said it looks sane enough; it's possible (and perhaps likely) that people did do that consideration.

But I think just washing your hands and saying "well, The Experts flunked too" is not a great defense from a rationalist perspective, even if it's an acceptable excuse from the world of Effective Altruism or the broader world of charity funding (or, uh, legal defenses). At the trivial level, rationalism is about being able to make useful predictions, and unless this was somewhere on your failure map (even at a low probability) or you absolutely didn't care, this points to some failure modes. At the very least, this should be a learning experience, if an expensive one.

((Especially from a Bayesian perspective. When it comes to big exchanges, we're not in none-pizza-with-left-beef days, here. If your prior for anything talking crypto doesn't include a lot of potential for fraud, you've not been paying attention.))

Maybe that's the meaning of #9, but it doesn't really read that way.

think about this stuff, with your own brain, in your own head ... Even if the 'only' problem with FTX was a hair-brained scheme that was simply going to go bankrupt without any fraud

But FTX's business made sense, they were a crypto exchange that made money off of crypto trading, why would they go bankrupt without fraud?

Crypto exchanges in general can be perfectly legitimate (uh, modulo legal compliance), and maybe even simple. You trade one currency for another, take a small cut, and run your business's operating expenses from that cut. You can add some complexities, like shifting where your assets are based on expected future rates, or the weird self-minted stablecoin, or reasonable loans, or what have you, and it's not automatically going to collapse. But there are variants of that business that don't make sense.

FTX specially has a problem: that "small cut". Crypto exchanges aren't easy, between legal compliance, potential risk, the difficulty of safe storage of assets, and the complex question of reputation. But they're also not monopolies, by their very nature. There's an upper limit to how much can be extracted from a given amount of assets, before you either get competition or your customers go broke. Not quite the efficient market hypothesis, but that there's a lot more one-dollar bills than hundred-dollar bills laying on the sidewalk. And while FTX's bankroll was big, its publicly-known expenditures were tremendous even for its size, often with large upkeep costs and likely hidden costs.

FTX claimed one billion USD revenue for 2021, based off... depending on how you measure it, 300-700 billion USD-equivalent in trades. That's high as the current market goes, but not unreasonable, not hugely surprising for a publicly reported number. (there's some flaws in this 'lump of labor cash' fallacy sense, since a lot of those USD-equivalent trades are in weakly or unconvertable coins, but they're not-obvious enough that I can understand people not seeing them without a deeper examination.) That would make an estimated two billion USD expenditures as double revenue the year before, and the /160 million USD funding already marked for grants by FTX Future Fund would be /6% of revenue. Note: that's revenue, not income. Either of these individually would be major warning signs.

There are rare cases where this level of discrepancy can make sense -- if the expenditures are one-offs, or if they're an important part of getting an very near rapid growth state, or both, or where the company is in early stages of development with little total income compared to its potential. But FTX was talking about scaling many of these costs up (to 1 billion USD in the Future Fund!). And even if FTX could keep its revenue-per-trade constant, a healthy income/revenue ratio would take something like 1.2-trillion to 2.8-trillion USD in crypto trades. Even if you think this is possible, or even more likely than not, it's at least something that can go tango uniform.

In a perfect world, a truly ethical company have these sort of problems and a proclaimed reserve requirement would fold without bankruptcy, but in practice there's a ton of things where an ethical company might not even have realized exactly the day they'd gone into the red (though it'd probably not be eight or twenty billion dollars late, either, so that's definitely better).

((This isn't specific to crypto or to charity; in the business world, nutjobs willing to sink a lot of someone else's money that they'll 'make up with scale' are more common than anyone needs.))