Unions probably would have gone a long way during all this.
this post was submitted on 25 May 2026
777 points (97.3% liked)
Programmer Humor
31576 readers
1601 users here now
Welcome to Programmer Humor!
This is a place where you can post jokes, memes, humor, etc. related to programming!
For sharing awful code theres also Programming Horror.
Rules
- Keep content in english
- No advertisements
- Posts must be related to programming or programmer topics
founded 2 years ago
MODERATORS
They’re all moving to tokens based billing to stop the bleeding. The markets aren’t as willing to let them burn infinite cash on operating expenses anymore. A lot of the private investors see the writing on the walls, that these AI companies are not worth anywhere near the valuations, and propping them up to ensure further datacenter build out to keep Nvidia shares high is costing more than it’s worth.
A lot of the model providers are lining up for ipos to cash out their initial investors and dump the responsibility of the mess on the public markets. They have to jack their prices to pretty up their books if they want to cash out with an IPO though. The unlimited access for a flat monthly cost shtick was obviously never a viable business model. They’ve all been spending massively more on inference than they’ve been taking in subscriptions, and even with the changes to token based billing, they still will be.
Anthropic is claiming that they will have their first “profitable” month soon, but that’s actually just a magic trick of book keeping. They’re deferring some cost of compute for a few months, and accepting a bunch of pre payments from customersr. Basically pretending to have high revenue and lower cost by shifting when the payments are made. And then on top of that they’re jacking up prices and thus decreasing their demand with the shift to tokens.
Everyone is running around rearranging deck chairs on the datacenter titanic hoping they can keep their feet dry a little longer.
Palantir calls for aid! Would Rohan answer?
You can't even properly review the result of burning 4.5k in tokens with just a single senior dev.
The real limit still is how much high-skilled natural neural compute you got to make sure that the AI output is actually any good.
Heaven's Door!
Open the server and write a query to restore the data!
Dude my nephew was telling me this an AI company galytix is literally running on Indian grads and claiming to be ai what a scam
AI: Actually Indians
I really hope they start poaching workers in outsourcing industry and make prices go up soon.
I really wonder to what degree “Self Driving Vehicles” are being completely or partially supervised and controlled by an overseas desk worker.
load more comments
(5 replies)
Literally was in a presentation at work the other day where they said someone used 2 million tokens and the next person used 1 million and they were so excited by it.
I didn't get it. I asked went 2 million was better than 1 million. The VP basically said it's not, just costs us more money.
So why the hell is the presentation acting like it's a fucking victory?
What is a token, anyway? I've never seen any mention of them with GPT.
Short answer is that a token equates to a unit of work. I strongly suspect that the definition has been becoming less and less tangible as the bubble has inflated though
Excellent question. I have absolutely NO idea. Friend helped me set one up on my PC and it even churns through "tokens" on my GPU. So I have zero clue. :D I guess I could ask "AI". lol
Tokens are what an LLM actually predicts, one after another. If you have very slow models that produce less than 5 tokens/s or so, you can easily follow it with your own eyes. A token is what appears at once. Often it is an entire word but it can also be parts of a word or individual letters, digits, special signs for uncommone words or special formatting, number stuff.
It's the ai industry's made up In game currency. That flat monthly fee they're trying to get rid of was basically the season pass
It's all fraud and gambling! :D
lol this is how I explained it to my wife. Like how arcades all have these money cards now.
basically the same mechanism. I'm heavily assuming they actually took inspo from the game industry
So gross. Obfuscation of currency so you have no clue how much you're actually spending. aka gambling imo.
Agree 100%
Because AI use is up! The FUTURE!
Because they believe that AI will replace payroll, which is generally a businesses highest expense. The more AI usage, the more likely they can eliminate jobs and give themselves bonuses.
They are excited the way a farmer is excited when his pigs are gorging themselves on grain and are getting really, really fat. The farmer is glad to pay for grain because its much cheaper than the money he makes from selling them for meat.
They are giddy because its almost time for payday.
payroll, which is generally a businesses highest expense
This is something that changed in the industry; it was a investment back then.
load more comments
(8 replies)
load more comments
(24 replies)
Because it's a proxy metric for adoption. Currently you generally want your engineers to be using it as much as you can afford to so that they all learn and improve and adapt existing processes etc
Which is a stupid mindset.
"Go forth and burn tokens and your performance will be measured on that"
Looks like I'm going to make a for to ask for a for every word in /usr/share/dict/words. Look at all the tokens I burned.
It doesn't reflect upon business value, performance, or education.
It's even worse than the disastrous lines of code metric.
Their problem is they have no idea what to expect, so to signal affinity to hype, they just measure tokens.
Also, I've been noticing more and more APIs are adopting a policy where they get rid of any unused credits that you've paid for at the end of the month, creating a 'use it or lose it' FOMO mentality.
All very normal and sane.
Engineers under such incentives should ask AI how to most easily and speedily consume as many credits as possible, I bet it knows a great way
In my experience with Claude Code Opus, analyzing or generating files eats the most tokens while thinking is actually surprisingly cheap. I guess, the token-counting is somewhat wrong on purpose to incentivize use of high-effort thinking mode because when you incentivize using lesser models or modes, people get disappointed by the output quality and stop using the service...
So just let it analyze the code base for flaws and bugs in a loop using lots of sub agents for each type of bug or code smell.
The good thing about that method is that it is technically malicious compliance; but it also offers a high degree of plausible deniability and likely yields some actual bug fixes to offer as justification.
Doing it just once every once in a while without fanning out into tons of agents rereading the same files is the non-malicious-compliance way of using AI for bug hunting and usually worth it. Also let it write tests for the found bugs (and properly review those tests using the natural neural network in your head).
Probably just ask it for the seahorse emoji or something idk
load more comments
(1 replies)
But why force adoption? If it's really good people would adopt it naturally like we do with every other tool. Why for this really expensive one, do we now flip the whole system on its head?
load more comments
(3 replies)
view more: next ›