This really sounds like an AI voice agent transcribed "Thirty Seconds to Mars" to "30 seconds for Bruno Mars" and then no one actually proof-read the thing.
I've see a post like this every week for the last 2 years. Are these models actually getting worse? Or do folks start noticing the cracks as they use them more and more?
I have to work with ISO-purchased pdf documents that are heavily DRM-controlled. I can only open them on two PCs with a plugin that only works in Acrobat reader. It is so closed and unusable.
Genuinely curious, what does that mean? Is the PDF standard now open and free, or does Adobe wield any power? Does it own trademarks (for pdf itself, not Acrobat Reader or sth like that)?
In theory, that means that anybody can create a fully compatible editor/reader, and if Adobe tries to change theirs it will be them that are incompatible.
In practice, I don't know why people are talking about PDF. In any recent time, I have only seen people using Adobe readers by accident and haven't heard about anybody using their editor for any reason. I know of some people that buy their editors, just not any that use it.
I occasionally use a few obscure Acrobat Pro features, mostly in the preflight tool with a UI that looks as though it hasn't been updated since the Clinton administration.
And I occasionally use Acrobat Distiller to convert old PostScript files to PDF, and to embed fonts into old PDF files created without embedded fonts, mostly because I already have it set up to find and embed almost every pre-OpenType outline font ever released by Adobe, Apple, or Microsoft, along with the OpenType fonts in Adobe Font Folio 11.1, and I've never gotten around to setting up all the fonts in Ghostscript.
The only Adobe product I use more than once or twice a month at this point is Photoshop, and mostly for things that almost certainly could just as easily be done in any number of other image editors at this point, or even ImageMagick. It's just that I've been using Photoshop since version 2 (not CS2 ca. 2005, Photoshop 2.0 ca. 1991), so it's comfortable.
It's been open and free on Adobe's website for 20+ years, aside from proprietary extensions like XFA. It's the ISO standards that until recently required payment (as that's how ISO generally works).
The market for used tractors went through the roof years ago--20 to 40 year old tractors with tens of thousands of miles on them sell for not so far from new prices because farmers value being able to fix them without paying $$$
This notion that "we don't have enough compute" does not cleanly reconcile with the fact that labs are burning cash faster than any cohort of companies in history.
If I am a grocery store that pays $1 for oranges and sells them for $0.50, I can't say, "I don't have enough oranges."
There is a major logic flaw in what you're saying.
'If I am a grocery store that pays $1 for oranges and sells them for $0.50, I can't say, "I don't have enough oranges."'
How about 'if I'm a grocery store and I see no limit on demand for oranges at $.50 but they are currently $1, I can say 'if oranges were cheaper I could sell orders of magnitude more of them'.
Buying oranges for $1 and selling for $0.5 is an investment into acquiring market share and customer relationships and a gamble on the price of oranges falling in the future.
Selling below cost is also called "predatory pricing". Sadly it's legal in US but it's something wealthy companies do to kill competitors and end up with captive customers.
The grocery store analogy works if compute is the orange.
But labs arent buying oranges — theyre buying the only orchard on the island, hoping it yields a fruit no ones grown yet. Burning $1B to net $500M isnt "I have too few oranges." Its "Im betting the farm Ill find a new one."
Both can be irrational. Theyre irrational in different ways.
"I built a ship to go to the Indies and bring back tea."
"Bro, the ship cost 100,000 pounds sterling and only brought back 50,000 pounds of tea. I don't care if you paid 12,500 pounds for the tea itself, you're losing money."
There is a very rational reason labs are spending everything they can get for more compute right now. The tea (inference) pays 60%+ margins. And that is rising. And that number is AFTER hyper scalars make their margins. There is an immense amount of profit floating around this system, and strategics at the edge believing they can build and control the demand through combined spend on training and inference in the proper ratios.
60%+ margins according to numbers which are not published publicly and have not AFAICT been audited.
Could they be accurate? Sure, I think people who claim this is impossible are overconfident. But I would encourage anyone who assumes they must be right to read a history of the Worldcom scandal. It's really quite easy for a person who wants to be making money (or an LLM who's been instructed to "run the accounts make no mistakes"!) to incorrectly categorize costs as capital investments when nobody's watching carefully.
Any materially false public statement by one of the foundation lab CEOs is a huge foot fault. I'm not saying they would never lie, but it would be a very, very dumb thing to do. That public information can be relied on by their private (very powerful) investors. I think if you're hearing these numbers ballparked in public settings, they are, as a prior, directionally accurate.
I agree, although I would emphasize that Worldcom is a great example of a CEO doing that very dumb thing. But I am not hearing these numbers ballparked in public settings. As far as I can tell, all the numbers people discuss for OpenAI or Anthropic margins come from anonymous leaks of internal documents.
- Selling those requests at less money than it cost to run the compute for those requests (because if you raise price clients go to openai)
The statements are not contradicting each other?
They keep subsidizing to try to grow customer base, but they can't serve the customer base they have, they're expecting customer base grows faster than it drops from people bothered with rate limits (it probably will, average user won't hit rate limits enough to change)
Probably expecting a breakthrough in efficiency for compute, or getting enough cash flow (IPO?) to get more compute before it all comes crashing down
Model inference compute over model lifetime is ~10x of model training compute now for major providers. Expected to climb as demand for AI inference rises.
reply