OpenAI has a negative 122% operating margin and growth of usage has stopped.

[pl@cosocial.ca]

@GossiTheDog nothing a few ads can't fix!

[fencepost@infosec.exchange]

@GossiTheDog oh no, will they be able to get enough investor money to pay for all the GPUs, RAM and storage they've committed to buying?

[awoogageneral@mastodo.neoliber.al]

@GossiTheDog absolute woof

[shtrom@piaille.fr]

@GossiTheDog

[scottmichaud@mastodon.gamedev.place]

@GossiTheDog -122% is actually better than I expect. I'm wondering how many "large line items" are being excluded. Does it include some things that are core to the business?

[chrisp@cyberplace.social]

@GossiTheDog

[dirkhh@hachyderm.io]

@scottmichaud
IIRC they are excluding the cost of training.
So this is just the cost of inference...

@GossiTheDog

[scottmichaud@mastodon.gamedev.place]

@dirkhh @GossiTheDog I hear training is frequently excluded in a lot of their numbers, but I don't know which ones. (And, of course, that is silly because their entire point is compressed human knowledge, so they're going to need to continually compress human knowledge.)

[azuaron@cyberpunk.lol]

@shtrom @GossiTheDog It would be absolutely amazing if the bubble popped before they even IPOed.

[dirkhh@hachyderm.io]

@scottmichaud
Also, of course, is the actual cost of obtaining training data. Because "stealing all the data in the world" isn't really valid accounting in most systems I'm aware of.

@GossiTheDog

[shtrom@piaille.fr]

@Azuaron Yeah! Though that would be no mean feat. IPO is when the VCs make their money back from retail investors and finally get to pull out. I suspect they'll sunk-cost until then @GossiTheDog

[scottmichaud@mastodon.gamedev.place]

@dirkhh @GossiTheDog So... well, in a utopia, it would be free. I can see them getting away with it being nearly-free in a dystopia. It could be super-expensive, though.

[gary_alderson@infosec.exchange]

@GossiTheDog the valuations seem to be basically fake - china is the value option now and they actually include freedom dust

[tootbrute@fedi.arkadi.one]

@GossiTheDog incredible business minds running AI.

[ggmcbg@mstdn.plus]

@GossiTheDog

Lie-writing. They didn't generate anything then.

[huronbikes@cyberplace.social]

@GossiTheDog really looking forward to the next batch of corporate financial thrillers such as "Margin Call 2: 2 Margin 2 Call" and "The Big Shart".

[cybeej@infosec.exchange]

@Azuaron @shtrom @GossiTheDog my over-under with my bestie is November. We’ve a fancy steak dinner riding on it. Remarkably just last week he admitted that’s it’s gonna pop, but he estimates Q1 ‘27 after Q4 numbers are cooked/announced.

[npars01@mstdn.social]

@shtrom @Azuaron @GossiTheDog

The last time petrostate despots & oil oligarchs laundered their cash via Deutsche Bank, private equity, & the housing market.

This time they're laundering their cash through US tech.

In any other situation, the C-suite would be swiftly turfed out, but Zuckerberg is still in charge at Meta, despite losing $77 billion on the Metaverse.

In the AI situation, the same insane losses & the same retention of money-losing executives. Odd.

Circular Finance Fraud.

[casandro@f-ckendehoelle.de]

@fencepost @GossiTheDog Yeah, that's why nobody is actually building factory capacity.

[richrants@toot.community]

@dirkhh Oh, if they moved from IFRS to AIFRS (which I just invented, stroke of genius, will need to start this as a side hustle) that’d be easy. They can just activate all the data they … found … as “Fair Use Data Inventory” at arbitrary valuation. Also, model training is not expensed, it’s an asset called “Capitalised Weights & Biases” (amortised over 99 years). And whatever they pay for compute = Deferred Compute Debt. Makes the numbers look so much better.

@scottmichaud @GossiTheDog