I am not a CEO and I hope this AI bubble bursts already.
That said, if I were a CEO using all possible tokens while they are heavily subsidized and tightening the purse when they get more expensive does not sound like the worst strategy to me. You get all your teams to build some expertise and (hopefully) get a sense of where the technology might have some ROI.
If only they had presented it this way (and not “AI therefore layoffs”) probably a lot of us would hate it much less now.
You get all your teams to build some expertise and (hopefully) get a sense of where the technology might have some ROI
You’d run the risk that they instead develop a dependency or overreliance on the tech. They probably won’t think about the “I” part of ROI and evaluate how many tokens a given task produces relative to the saved time and effort.
Throttling it later might then cause a drop in productivity until they relearn how to do simple stuff they could do themselves but delegated to AI instead, whether or not it’s ideal for the task.
For example: “search and replace” requires the LLM to ingest and then produce the whole document as output. Aside from the question whether it’ll have caught all instances and replaced them without otherwise altering the text (which a casual user won’t check), the amount of output tokens correlates with the size of the text.
That’s a lot of wasted tokens for a task they could have done without AI, but so long as asking the computer is quick and convenient, they won’t think twice. Then, once the tokens are throttled, they’ll suddenly realise they’ve run out of tokens early because they burned a ton on tasks that seem trivial to them, leaving none for the more complex tasks they’d actually prefer to delegate (whether or not they should). They might not make the immediate connection which tasks eat so many tokens either, so they’ll take a while having to try all their use cases again, see how expensive they are, run out of their allotment early and wait for the next period.
If you’re gonna have people figure out how to use it, you’ll have to throttle from the start to make them also figure out how to use it economically.
Also, mandatory classes on the limitations and reasonable uses. Don’t let it get to the point where they find out the hard way that it’s not actually intelligent and has no concept of truth.
I hadn’t thought about this, thanks. Personally, if the messaging had been “play with this new thing, see where it helps, report where it doesn’t or where it’s actively harmful”, I would have had a much better time with it. The fact that it was “use AI for everything or else you’ll lose your job to someone else who does” created all sort of perverse incentives to use it for the sake of using it (even where it doesn’t make sense), to lie about the results and to generate more anxiety in others to keep up with your made-up achievements. I think at least some of the wasteful or even harmful ways you describe of using LLMs come from this push to use it and “be more productive” with it.
But you’re right that there are people who became overly reliant and even ruined their lives with LLMs without the tech being forced on them.
to lie about the results and to generate more anxiety in others to keep up with your made-up achievements
Emperor’s New Clothes style “we all need to pretend to like it” is an unfortunately common effect of decision-makers deciding they know some brilliant thing and any naysayers just aren’t suited to appreciate the brilliant thing.
I think at least some of the wasteful or even harmful ways you describe of using LLMs come from this push to use it and “be more productive” with it.
Some, sure.
Others from a fundamental misunderstanding of the nature of language models. They’re text processors and generators designed to sound human. They can’t tell facts from filler.
Just earlier, I saw a post elsewhere about someone having generated an article or something which cited three experts – wrongly, because it doesn’t actually know what the relation between the text in quotes and some supposed source is or why it needs to be verbatim to be a correct quote. That’s not a bug, nor a hallucination or whatever anthropomorphic euphemism people come up with for “random output happened to be wrong” (though, to be fair, “random” glosses over a highly complex prediction system that can predict plausible text quite impressively, even if it can’t predict truth).
Students relying LLMs to generate their coursework are falling into that trap without any pressure of productivity. They don’t get that the purpose of coursework is to learn about the source material and the structure of academic writing rather than just produce text. They also don’t get that the LLM won’t look up, interpret and cite sources accurately in accordance with the subject of the question. It will generate a plausible-sounding answer to the question, and therein lies the danger: If you don’t already know the answer, how could you tell if it’s true?
The same goes with people “looking up” information. Gemini will produce some text statistically correlated to the text it has read, but you never know whether that correlation reflects facts or whether it falsely attributes some shady business to companies who had nothing to do with it (about which there was a court case in Germany recently).
Vibe coders without programming experience cannot qualify the output of their generator. It’s always harder to understand code you didn’t write (or maybe wrote long ago), but if you don’t even know how to write code, you’ll have no experience to compare it to.
People using AI for coping with stress may run into a trap where they end up unlearning to cope on their own and potentially take on even more stress.
The common thread behind these is that these AIs lack the understanding of the concepts they’re producing text about and semantic connections between them, and accordingly cannot treat these things with the same nuance and precision that humans can.
But the ways they’re harmful doesn’t immediately become apparent. “Report where it’s harmful” doesn’t really work if it takes two years for a critical security flaw to surface that some code generator produced and nobody with experience caught. You may never notice your ability to deal with stress being eroded until some day you can’t ask your robot buddy for help and just crack instead.
They plant traps in your education, your knowledge, your work, your psyche. To encourage people to use them without thoroughly preparing them for those traps is reckless.
There is a limit to how long an AI company can keep subsidizing the tokens, eventually the financial ruin ensues. Every chat request that goes to the LLM has an actual physical compute and RAM cost, which scales to hilarious levels as you keep chatting in the same thread and the context size widens. It scales to astronomical levels when the query requires high-level analytical or reasoning skills like
thinking...,pondering...,bloviating..., etc. - which is exactly what enterprise users intend on doing. The Uber story made it quite clear - even some of the big techs don’t have the stomach for that kind of unlimited resource drain.
So the the test of capitalism is whether any of the executives who pushed the AI roll out have gotten fired or had their bonuses slashed… My guess is none of them.
If this were actually about competition, then people would be punished for not paying attention to all of the naysayers who predicted this exact phenomenon. If accountability were a key feature, then corporations would have set up their bonus structure to look for 5-year or 10-year benefits from the AI push because of this exact issue.
Of course we haven’t seen that anywhere because AI was and always is a bubble and everybody knew it and the only goal was short-term profits for whoever can claw them out of the employees or the minor shareholders fingers.
This, plus if we had any kind of political will or intelligence (as a nation; meaning the USA) we’d force AI companies to pay their extenalities: treat and sanitize every drop of water they use, and build the infrastructure to bring it back to communities; pay for their electrical infrastructure in advance and pay their electricity bills to the tune of “nobody else’s bill goes up”; some kind of massive carbon capture tax for their use (this one might not be possible to actually do; it’s too much); and of course paying royalties and copyright violation fines.
As at least one AI CEO has said, if they had to pay for all the laws they’ve broken and resources they’ve stolen, all AI companies currently existing would go out of business. let’s say they didn’t: The cost per token would be quite high, and very few people would use it.
It runs on theft and planet-scale destruction.
It runs on theft and planet-scale destruction.
You could argue that this is the very nature of capitalism: theft because it always means owners extracting value from other people’s work, and ultimately planet-scale destruction because it depends on infinite growth while externalizing (not paying for) the true costs of its activity.
In that sense, AI companies are just a faster-growing strain of the global cancer that is capitalism.
Reasonable take, but the last part (“faster-growing”) is huge here. The sheer scale and multipliers on the bad things caused by AI are far beyond most (all?) previous technologies used by capitalists.
As at least one AI CEO has said, if they had to pay for all the laws they’ve broken and resources they’ve stolen, all AI companies currently existing would go out of business.
I find those terms acceptable.
It will only get much worse.
What’s the most expensive thing one can do with AI and how do we do it on a mass scale to one company at at time?
I know a company that burned $100k in tokens after they they let like 50 worker bees using general AI for OCR, simply converting images and PDFs to text.
They didn’t bother to create a skill, or teach the AI how to reuse a shared script so every request resulted in it writing a new python project, pulling libraries, using a frontier model rather than offloading a dumb one etc.
Basically find a business process that happens often and let em at it inefficiently, it’ll happily chew through the budget.
Thats pretty much what people freaked out about llms doing at my work and all they use it for. I’m here like…we have had OCR for over 20 years.
People are duuumb.
There has been some serious leaps in terms of quality. It couldn’t read human writing or half the fonts for that matter like 5 years ago, let alone 20.
OCR libraries have undoubtedly improved but LLMs are using the same open source libraries and tools available to anyone… there’s few cases where sending the work through general models is worth it for text conversion. Employees just needed a front end to upload, run something like tesseract behind the scenes, and spit out the result. It’s an egregiously stupid use of resources.
have undoubtedly improved but LLMs are using the same open source libraries and tools available to anyone…
I read a surprising article on Lemmy just a week ago that explained that that is not how LLM’s do OCR. LLM’s convert images into tokens and then treat them like text input. I can’t see how it works but it does. It’s why they are better than classic OCR neural nets but at the trade off of enormously larger computation cost.
deleted by creator
Maybe 16k image/video upscaling of super low quality shitposts?
File sizes are going to be huge! 2K is already a lot to upload, couldn’t imagine 16K right now.
Just expend some more compute time on doing compression and we’ll get those filesize numbers to a workable level.
$ stat -c %s enhance.png 276773 $ convert enhance.png enhance.avif $ identify enhance.avif enhance.avif AVIF 1164x558 1164x558+0+0 8-bit sRGB 14391B 0.000u 0:00.000 $ stat -c %s enhance.avif 14391 $
zoom and enhance
$ identify enhance2x.avif enhance2x.avif AVIF 2328x1120 2328x1120+0+0 8-bit sRGB 32448B 0.010u 0:00.000 $ stat -c %s enhance2x.avif 32448 $
zoom and enhance
$ identify enhance4x.avif enhance4x.avif AVIF 4656x2232 4656x2232+0+0 8-bit sRGB 50758B 0.000u 0:00.000 $ stat -c %s enhance4x.avif 50758 $
Okay, that last one took 17 minutes to upscale on my GPU, so I’m not going further. But I’m using SD Ultimate Upscale, which is tile-based, so in theory that could be farmed out over a collection of GPUs and parallelized. Just need more compute hardware.
But as to filesize, that’s under 50kiB.
Whoah! Nice effort! Yeah a cluster could cut the time down a lot.
Burn your AI tokens and get promoted for using more AI!
I sadly don’t find the GitHub repo with the bash script anymore, but that thing seems to do the same thing
(Maybe I’ve misread though, I’m just going to sleep)Edit: ha! Found it!
https://github.com/dtnewman/burn-baby-burnhave it end to end build a fully featured web browser that works on Windows, MacOS, and Linux from scratch.
Thus transferring their money to openAI, Anthropic etc? How does that help?
They actually lose money as well, because they’re proving those tokens at below cost price.
It’s all going to Nvidia
those companies arent profitable either and they have same problems in which it costs them more to run their products than they are currently charging people to use it.
What do you mean by either? Walmart and Amazon make tens of billions in profit a year if not a quarter.
i thought we were talking about OpenAI, antrhropic, etc, not Walmart?
Did you read the title? It says to spend Walmart and Amazon’s money on AI. And you said “those companies aren’t profitable either” which would mean, using normal rules of English grammar, that “Walmart and Amazon aren’t profitable and OpenAI and Anthropic aren’t profitable either”. So what are you talking about? What does “either” mean?
ooh I see. my bad. erase the word either from my comment, it wasn’t necessary.
More companies with less money is better than a few companies with all the money.
Ultimately distributed power has to be more democratic, and centralized power has to be more fascistic.
That’s part of why governments having large distributed bureaucracies each with their own authority and independent ability to intervene is better than say; a single executive office/president controlling everything directly.
Distribution also leads to stability though (making it harder to challenge the status quo), so it’s a double edged sword.
“The public really didn’t like it when we shoved AI down their throats. Whats the solution?” “How bout a lil’ ‘artificial scarcity?’”
Removed by mod
Article text?
€69 per month
Complete digital access to quality FT journalism with expert analysis from industry leaders. Pay a year upfront and save 20%.
deleted by creator
