The demand for AI simply doesn't exist at the real prices. It barely exists at the current subsidized rates - Microsoft, Google, Meta, Anthropic, OpenAI are spending hundreds of billions to make mere billions.
And then these data centers will be worthless, future ones won't get built, memory demand will evaporate on the spot.
Very interesting and thoughtful comment. I agree. When demand from hyperscalars drops rapidly after the tide goes out and everyone realizes that they can't continue building out the data centers, then fabs will be left with overcapacity that will flood the market. Wondering what this will mean for local LLMs when good RAM is available for cheap?
I'm yet to see any convincing argument that inference is subsidised in any substantial way. Training and speculative expansion are where the spend is from what I can see.
A few days ago Gemini redid their rate limits, making images/audio/video generation much more expensive, shrunk limits across the board (including a new weekly limit) and added more expensive tiers.
At the moment you can pay $20/month to do thousands of expensive queries a month (involving file uploads, the Pro model, extended thinking), and evidence suggests that heavy users are not profitable.
I agree that heavy users are probably not profitable but that's the way the economics of subscription services tends to work across the board.
I'm arguing that even if inference isn't profitable right now it's not orders of magnitude off. Whatever pricing emerges for models equivalent to current frontier models won't be significantly higher than the current API pricing.
There are already enough small companies without tons of VC money to burn that are serving up nearly-frontier llms at prices lower than the big players are charging. They can't all be subsidising? These are companies without any moat or any IP.
If inference was profitable - they'd tell us. Msft, goog, public companies. They'd break out the numbers and show us, if they were good.
But instead, all we get is known liars going on podcasts and repeating "stylized facts" that aren't literally true about their supposed profitability on inference, from companies losing billions per year in a situation where they don't have to tell the truth.
That is VERY far from a convincing argument that they are profitable. So I can & will safely conclude that the opposite is true.
Nowhere since gpt wrote it. Normal humans dont talk like that. Normal actual humans with veins, arteries and a fully functioning endocrine system talk like ME
This article is written atrociously. It's geniunely unpleasant to read the basement tier slop like this. It doesn't matter if it has a salient point because this article could have been a comment.
Agreed, I can't read articles anymore with that Claudish stench of confidence, impactful sentences and dramatic negative parallelism. At some point you just start to skim it and reverse engineer the simple original prompt which is "write an article about how people don't speak up because it can hurt their career, use real world examples".
I just realized what the "AI summarize" feature is for. This kinda turd. I'd never wanted to use it, because either I wanted the joy of reading whatever it was - or needed the details in full because it's documentation. Now, finally, a use. Manmade horrors of daily life.
Remember sharing simple thoughts between each other? Now you can do the same thing but with two AIs in between, one that produces words out of an idea and one that collapses them back. This is the way of the future.
Actually you don’t need an AI slop generator for processing AI slop when the headline is right there and you have more than two brain cells (presuming of course you are not a bot.)
> The information existed. It just never traveled. Nokia sold its phone division in 2013 for $7.2 billion.
(Incoherence)
> It's not saying "this is wrong" — that doesn't work. Real pushback is making problems visible: putting a price on the decision, naming what can go wrong, making the tradeoffs concrete.
(Excessive vagueness, and no awareness of irony)
> It's not that nobody knew. Everyone knew. Speaking up just wasn't a rational choice.
There's more that's off, and I "enjoyed" reading that about as much as being molested by the large language monkeys that apparently created this text. (Because it is text - mere sequential characters with no value - rather than an expression of useful thought.)
You could also think of this as "AI is the toddler that's eaten crayons and shit on the wall." Sometimes the results are funny and worth sharing, but an adult creative director is required.
Here it seems the toddlers have acquired brethren and no adults are to be found.
> Why do they think they are "helping" with hallucinated rubbish that can't even build?
Because they can't tell the difference between what the machine is outputting, and what people have built. All they see is the superficial resemblance (long lines of incomprehensbile code) and the reward that the people writing the code have got, and want that reward too.
Define "average" and "very good" - it's quite easy to become good enough to beat all your friends and family (as long as you haven't made friends at the chess club or chess competitions). But if you want to do your best at the local chess competition held in a school hall at the weekend against all kinds of people, from little kids to pensioners, then yeah, you're going to need to spend lots of time studying openings, learning end game theory, and solving chess puzzles.
strongly disagree that studying openings is necessary to "do your best" at competitions. In my experience almost all games between players under 2000 (class players) are decided tactically. I'm expertish (2200+ bullet, 2200+ blitz, 1900+ USCF, win most local tournaments in my area etc) and I don't bother studying openings. Chess is 99.9% tactics at the class level. You won't reach GM without opening theory memorization but you wont reach GM anyway.
Also a reminder for anyone reading these comments that chess should be fun! Don't let psychological hangups like thinking u need a good memory, thinking you need to study openings, have a certain level of skill, or need to play a certain format (like avoiding blitz because it is "bad" for your game or thinking OTB is more important) stop you from playing chess! The only rules for how to play chess are the rules of the game; all the other stuff e.g advice about how to get good are just things people make up. Learn and play however you want and in whatever way brings you the most joy! Chess is a game and it is meant to be fun and not be taken seriously
It doesn't take extreme memory on your part to remember to avoid that opening after the first 9 losses, or indeed the first one. There are 5-10 other reasonable options for you on the first move alone.
It doesn't take extreme memory on your friend's part either if you keep falling for the same trick. It would take extreme memory for him to have something prepared against every plausible option you could choose.
There is definitely value in that, but that value is outweighed - dominated, even - by the incentive produced to fix outcomes, incentivized by that money put on the line.
The demand for AI simply doesn't exist at the real prices. It barely exists at the current subsidized rates - Microsoft, Google, Meta, Anthropic, OpenAI are spending hundreds of billions to make mere billions.
And then these data centers will be worthless, future ones won't get built, memory demand will evaporate on the spot.
reply