Angry devs vow to flee GitHub Copilot as metered billing takes hold

Developers seem to hate Microsoft’s new usage-based billing policy for GitHub Copilot as they report burning through a month's worth of credits in hours. “This is a staggering shift from a 'predictable subscription' to a 'stressful meter-based' service that hinders my productivity rather than helping it,” wrote one developer on …

[Ace2]

“Gone. For basically nothing.”

AI in a nutshell.

$WORK is pounding us hard to use as much as possible… I’m eagerly awaiting someone to notice the bill.

[Anonymous Coward]

Re: “Gone. For basically nothing.”

Just make sure you have something to show for it, even if its an empty source tree.

As an aside, I wonder what Dave Plummer will have to say about his experiences with this new billing model. Maybe I should ask in that other trip-hammer microsft billing article..

[Anonymous Coward]

Re: I wonder what Dave Plummer will have to say

Fuck Off AC. Post with a handle. Especially when calling people out by name.

[Phil O'Sophical]

Re: I wonder what Dave Plummer will have to say

Says someone posting as AC...

[sabroni]

Re: Says someone posting as AC...

I always reply as AC to ACs, I reply with my handle to users that post with handles.

You've got to have a system.

[Jou (Mxyzptlk)]

Re: I wonder what Dave Plummer will have to say

Fuck Off AC. Post with a handle. Especially when calling Fuck Off.

[sabroni]

Re: Fuck Off AC. Post with a handle. Especially when calling Fuck Off.

I post with a handle when replying to posts with a handle. I post AC when replying to ACs.

I will not fuck off.

[James O'Shea]

Re: Fuck Off AC. Post with a handle. Especially when calling Fuck Off.

You will, however, fuck up.

[theOtherJT]

Re: I wonder what Dave Plummer will have to say

When someone has a youtube channel with over a million subscribers where they post videos showing their face using their real name I think it's fair to assume they're OK with the wider world knowing who they are and being referred to by name no?

[sabroni]

Re: When someone has a youtube channel with over a million subscribers

You miss my point. AC is for when you have something business sensitive to post, have a valid reason for hiding your handle or are replying to another AC.

Why would you need to hide your handle to call out someone by name? Who was it that wanted to finger Dave whatever his name is but couldn't let us know who they were?

When someone hides their handle for no good reason it makes me wonder why.

Nothing wrong with a bit of curiosity, especially with all the shit stirrers around at the moment.

[Elongated Muskrat]

Re: When someone has a youtube channel with over a million subscribers

AC is for when you have something business sensitive to post, have a valid reason for hiding your handle or are replying to another AC

And these are rules that are written down, yes? That can be verified to be accurate and complete? Or, perhaps, are just your personal set of rules for posting AC.

I can think of any number of reasons why someone might post AC other than the ones you have listed, for example, if someone wants to hide that they are posting on a forum when they should be working, but otherwise usually post with a recognisable handle - just to pull a completely random example from the ether.

I don't see why you'd need to be posting as AC to respond to another AC, either. What is the logic in this? If I have something to add to what another poster has said, whether they posted as AC is usually completely irrelevant to what I have to say.

As a rule, I post as AC where my post might identify me, my employer, or someone else whose privacy should be protected. Sometimes I might post as AC for other reasons, such as where the person I am replying to has known beef with me (hello various RW nut-job trolls), and I would like my post to be read by others without being immediately swarmed with downvotes, etc. I can anticipate that other people might have their own reasons for posting anonymously, to which I am not privy. This isn't to say that posting anonymously shouldn't arouse a certain amount of suspicion where motives are concerned.

In any case, it's pretty bold for anyone to dictate to others what the ability to post anonymously is or is not for. That is what Sir Humphrey Appleby might describe as a "courageous decision".

[MrBanana]

Re: When someone has a youtube channel with over a million subscribers

I only occasionally post as AC, mostly because I can't remember my password and I'm using a system that isn't my own, and doesn't have my El Reg login details cached.

[abend0c4]

Re: “Gone. For basically nothing.”

I’m eagerly awaiting someone to notice the bill

When they do, it will be your fault for being profligate. The whole AI delusion is based on the notion that staff are the problem and their elimination is the answer. Challenges to the delusion will be treated as sabotage.

[Anonymous Coward]

Re: “Gone. For basically nothing.”

"The whole AI delusion is based on the notion that staff are the problem and their elimination is the answer."

How long will we acknowledge that people are the problem but fail to accept that the Terminators^W....AI...are the solution?

And before you suggest that people aren't the problem, take a look at the people that surround you or switch on daytime television...sure, there might be a couple that just need a minor cattle prodding rather than the 'full service" but that's just a handful of exceptions.

[drankinatty]

Re: “Gone. For basically nothing.”

Well put. And from the opposite angle, the predatory billing would make any good drug dealer blush. First hit is always free ... they always find a way to pay for the next one. Hype or crutch, they've got programmers by the short-hairs with the coding models and assistants. The free hits have all been given away.

The third side to the fleecing is the numbers have to support the investment somehow. Though MS likely has the least exposure in the circular-financed AI brotherhood, they have nearly 40 years experience extracting what the market will bare. Somebody has to pay for the azure-credit skin they have in the game. For the rest of the brotherhood, perpetual fundraising isn't a business model, that is likely to be a hard lesson to learn.

[Benegesserict Cumbersomberbatch]

Re: “Gone. For basically nothing.”

I also love the tacit acknowledgement that bait and switch was the business model from the start.

Putting in a public announcement words to the effect of "you customers are only just now paying for the actual cost of development and operation of that tool you've been using for the last n years" says openly that enshittification was planned. Blind Freddie could have seen that, though.

Problem is that the people who made the decision to integrate an enshittified business model into their own plans now have to be the ones called on to notice that, hey, the way we did things before we made that choice they advocated was cheaper. Anyone... Anyone...

[NoneSuch]

Thank you Goodfellas....

Now the developer’s got Copilot as a partner.

Any problems, he goes to Copilot.

Trouble with the function? He can go to Copilot.

Trouble with the build, the tests, the docs, the pull request, he can call Copilot.

But now the guy’s gotta come up with Microsoft’s credits every month, no matter what.

Project bad?

“F*** you, pay me.”

Oh, the agent looped for an hour?

“F*** you, pay me.”

Copilot rewrote half the repo and broke the other half, huh?

“F*** you, pay me.”

Also, Copilot could do anything.

Especially run up usage on the account’s credit.

And why not?

Nobody’s watching the meter anyway.

And as soon as the prompts go in the front door, the credits go out the back.

You take a twenty-dollar subscription and you burn it down in an afternoon.

It doesn’t matter.

It’s all billable.

And then finally, when there’s nothing left, when you can’t run another model, ask another question, or let the agent touch another file, they bust the plan out.

They own the meter.

[NoneSuch]

Re: “Gone. For basically nothing.”

You don't understand.

Microsoft need to charge devs that extra money to not pay out bug bounties. So your money is just resting in their account.

[Rogerborg 2.0]

Re: “Gone. For basically nothing.”

Correct. I got the "how very DARE you be using AI, just because we told you to use AI" email yesterday.

[Dan 55]

Re: “Gone. For basically nothing.”

I’m eagerly awaiting someone to notice the bill.

Remember to run a script in the background to make it iterate through your source code and carefully review each file, no mistakes.

[tiggity]

Re: “Gone. For basically nothing.”

As someone who (for work purposes unfortunately has to occasionally use various "AI" tools (though long since stopped using GitHub copilot as it was not that great) I can say that performance on coding tasks is very variable - can be surprisingly good, can be awful. The problem is that, when pricing starts to get more "realistic" (i.e. companies want t start making money) that you can burn a lot of tokens (cash) for an awful "solution".

Until the "solution" an "AI" provides is consistently OK / good people will get very upset at burning cash for a dismal "solution".

With the relatively subsidised pricing then if "AI" crashes & burns in terms of a good solution then people will alter prompts and try again hoping for a better outcome based on what the "AI" got wrong last time.

With "realistic" pricing needs to be some way to credit users when "AI" just fails to do a good job as when more outlay is involved then people will demand consistently good results (what's the point of burning tokens for no useful code being produced?)

"AI" coding tools are getting better, but (obviously I have not used all of them, but based on what I have used ) are not yet at the level you can rely on. Though they are now a lot better at relatively small, very well defined tasks, which is (IMHO) best way to use them e.g. usually very good at writing your unit tests (including lots of edge cases if you prompt nicely) based on interface definition and expected behaviour specification for TDD.

[Anonymous Coward]

Re: “Gone. For basically nothing.”

I read a good thread on why using LLMs to create unit tests is not as good as it seems at first. I have to say he makes some good points and I'm now having second thoughts.

You can argue that AI generated unit tests are better than no tests at all, but what if the tests are actually not testing what you should be testing and give you a false sense of security?

[Scotthva5]

The first one is free...

Did they really think "all you can eat" AI was going to be around forever? Someone has to pay for the vast sums being wasted on the "AI" boondoggle and it sure as hell won't be Microsoft.

[Jimjam3]

Re: The first one is free...

Agreed, no such thing as a free lunch.

Colour me surprised !

[Headley_Grange]

Re: The first one is free...

The first one isn't free - it's an advance.

[Kraft]

Re: The first one is free...

People shouldn't. That's how mafia business has thrived : provide product for free until hooked, then propose a deal your user can't refuse.

[Alumoi]

Re: The first one is free...

Mafia, companies, governments, is there a difference?

[Pascal Monett]

Unfortunately, not any more.

Except for one thing : with the Mafia at least, if you do what they say, they will honor your obedience.

[Ochib]

Re: The first one is free...

And Games Workshop. The first Space Marine is free and they will help you paint it. After that it's bend over and grab your ankles

[retiredFool]

Re: The first one is free...

I think you mean drug dealers. Mob has always used more of a "thats a nice store you have there, pity if it burned down", "we can offer protection against vandals starting it on fire". And then there is also just the, "we will beat you up/kill you unless you pay" model they use. Drug dealers on the other hand get you hooked on a few free hits, and then start charging once you have to have it. Problem I see for AI is I'm not sure they will ever have the "have to have" and I don't expect them to openly start threatening concrete galoshes if you stop.

Now they will make it nigh impossible to cancel your subscription though. SOP for tech co's.

[MachDiamond]

Re: The first one is free...

"Did they really think "all you can eat" AI was going to be around forever?"

Some do. It's more that is goes from free to hugely expensive in one bite. Get a company hooked, let them dismiss a load of staff now that the new tool is working great and then lower the boom. Those fired people aren't coming back so the company has no choice but to suck it up and buy a bigger token package.

[katrinab]

Re: The first one is free...

You pay a subscription for Excel and you can do as many formulae as you like.

People would expect CoPilot to work in the same way.

[abend0c4]

Re: The first one is free...

You pay a subscription for Excel and you can do as many formulae as you like.

People used to buy Excel outright. Every financial model can be adjusted to maximise return.

[Elongated Muskrat]

Re: The first one is free...

I could probably be persuaded to pay a subscription so I didn't have to use Excel, if they could guarantee it.

[Eric 9001]

Re: The first one is free...

It was never possible to buy Excel - all that was ever previously available for sale was a license to use, that could be revoked manually.

The only difference now is that the license to use is automatically revoked, via the software refusing to function, after a set time period, with another payment required to extend the functioning time (also, current "subscription", but for some reason login isn't working? Too bad.).

[Kurgan]

Re: The first one is free...

You pay a subscription for Excel and you can do as many formulae as you like.

For now. But wait for the new pricing model.

[Juha Meriluoto]

Re: The first one is free...

Now that's an idea... Pay-per-formula Excel! Well, most people use it like a paper notepad anyway...

[big_D]

Re: The first one is free...

The wrong analogy. You pay to run Excel on your local computer, so your local computer and local electricity bill is taking care of that load. With AIaaS, the load is running on somebody else's computers and they cost a fortune and cost a fortune to run. At most, with M365, you are paying to have the files stored in Microsoft's cloud, and you get a fixed sized quota that you can store, if you want more, you have to pay more. But it is a known quantity, you aren't paying Microsoft for 1TB of storage, but you are actually storing 30-40TB a month...

The other thing is, Microsoft sells 1TB to each customer, but 99% of them only use a few GB, so they don't need to have 100% coverage for storage on day 1, they can build it up and the low usage users are subsidising the high usage customers. When it gets to a certain size, they have to raise prices, because they need to invest in more storage and more quantifiable compute power for the web services.

But, on the AI front, even the most lightweight user is using more compute resources than their subscription pays for, so there is no unused quota that can be put to high usage users to offset them. EVERY user is losing them money. Now people have become used to using it, they are now being asked to pay for what they use...

[Groo The Wanderer - A Canuck]

Re: The first one is free...

This. 110% this.

Even companies with actual deep pockets, like Microsoft, and unlike the pyramid scams known as OpenAI or Anthropic that are funded entirely by investments, are starting to feel the demand from their investor base to answer the question: "Where's my ROI?"

Artificial Ignorance service provisioning is no different than any other business; costs have to be covered. And that processing hardware is not cheap, nor is creating the models, and nor is running that processing hardware when you're looking at such scope and scale.

I expect most "personal developers" are going to have to give up on the current crop of AI tools; they're about to become much too expensive for personal use. Time to learn how to leverage a local model and put up with it's limitations and start doing your own damned work again!

[Elongated Muskrat]

Re: The first one is free...

There's another option that isn't "use your own local model". It's to learn and develop the skills required for proper software development yourself, so you can design, code, and maintain your own software yourself, without having to pay obeisance at the temple of "AI".

[Groo The Wanderer - A Canuck]

Re: The first one is free...

Hence "start doing your own damned work again."

[big_D]

Re: The first one is free...

Exactly, these companies have been racing to the bottom by offering their services at a loss, trying to gain customers, now they need to actually cover their costs and people complain... Go figure.

We have known that these companies are losing money on every query made and that at some point they would have to actually start charging real money for these services.

Looking at the technology, it is still developing rapidly and the billions invested in hardware this year will need to be re-invested in the next generation in a few months, because the next generation is faster and more efficient... Didn't Microsoft have a warehouse full of nVidia cards they didn't have power to drive and by the time they got the power, the chips were so outdated (6 months), that it was cheaper to write-off the cards and buy new ones?

Somebody has to pay for this madness, and it shouldn't be the non-AI using customers of these services subsidising the heavy AI users. The AI pioneers have known they were getting a lot of usage for pretty much nothing, now the real bills are starting to arrive and they are "outraged" that they have to actually pay their way. If the AI is making their job easier and more efficient, they should be saving enough time and effort that the AI still pays for itself. If it doesn't, it is the wrong tool for the job and they should look for a more suitable tool that is more cost effective.

[that one in the corner]

The well-known name ranking

Pro < Pro+ < Max ? Max?

Oh, that wasn't the important point about the story? You mean that stuff about Microsoft putting up the price? Or everyone threatening to walk away from the service? A service that is so useful that it delivers “It gave some pretty mediocre suggestions. Didn't really solve the problem, I still had to do most of the work myself"? How is that the important part, nobody can be even slightly surprised about any of that!

Anyway, I was hoping after "Pro+" (isn't that something to do with Ibuprofen?) they'd do a subscription plan where, for the extra fee, they let the monthly limits be lapsed for a rollover period and call this (you are way ahead of me now, aren't you!) "Copilot Prolapsed".

[ComputerSays_noAbsolutelyNo]

Re: The well-known name ranking

Pro Max Special?

Max Pro Ultra?

... never underestimate the Marketing Department

[news.bot.5543]

Re: The well-known name ranking

No, because I can top yours...

Super Ultra Max Pro: Special Limited Edition

Comes with a free pint, so you don't notice the extra zeros added to the monthly fee

[David 132]

Re: The well-known name ranking

Super Ultra Max Pro: Special Limited Edition

You forgot to append "Turbo Classic Enterprise". That's the sku that the real enthusiasts want.

[I ain't Spartacus]

Re: The well-known name ranking

I genuinely saw (I wasn't paying the price to buy) Finish Quantum Ultimate dishwasher tablets - with a yellow label on the corner saying "New Improved Formula". How do you improve on ultimate? And is that a threat? I guess improving on Quantum is quite easy - but you don't know if it's worked until you open the box...

[BartyFartsLast]

Re: The well-known name ranking

Well, to be fair, I'm always quite surprised at how good a job the dishwasher does, so I guess "quantum" is appropriate

[I ain't Spartacus]

Re: The well-known name ranking

Well, to be fair, I'm always quite surprised at how good a job the dishwasher does, so I guess "quantum" is appropriate

It's not all that quantum. Every time run my dishwasher with a cat in it, when I open the door the cat is dead. I thought it was supposed to be a 50/50 chance the cat will live?

Bloody Schroedinger! I'll tell you I'm not buying one of these expensive German dishwashers again! Next time, I'm getting a Hotpoint.

[Phil O'Sophical]

Re: The well-known name ranking

It's like those detergents to protect the colour of dark clothes. They remove the whitening agents, and then charge more for the "special" feature.

[Elongated Muskrat]

Re: The well-known name ranking

My personal favourite is the current heavily advertised product that is a double-sized washing "tablet" (I use quotes, because these things seem to have evolved into some sort of jelly-filled pouch) marketed as being able to deal with "tougher" washes. It seems that the people who buy this sort of thing are so thick that they couldn't otherwise count to two, and, you know, just use two normal ones.

[Peter Gathercole]

Re: The well-known name ranking

I always find it amusing when they mix "original" and "improved" in the same set of superlatives.

[seldom]

Re: The well-known name ranking

With added Quantum and Blockchain

[Joe W]

Re: The well-known name ranking

Nah, they'll lower the number on the pricetag and change it from "monthly" to "weekly"...

[Mark #255]

Re: The well-known name ranking

That's rather optimistic, believing that the price rise is going to be less than 4x

[Charlie Clark]

Re: The well-known name ranking

A new variant on shrinkflation?

[An_Old_Dog]

Super Ultra Max Pro: Special Limited Edition

Does it also come with a Special Limited Edition Master Shooter Master Sergeant Gamer Helmet you can wear?

[Guido Esperanto]

Re: Super Ultra Max Pro: Special Limited Edition

No sadly.

For that you need the "Super Ultra Max Pro: Special Limited Edition: Remastered" version

[breakfast]

Re: The well-known name ranking

The challenge for marketing departments here is that most of these names are already in use for sanitary pads.

[MrBanana]

Re: The well-known name ranking

Sounds like it came from the marketing dept for a feminine hygiene product.

[James 139]

Re: The well-known name ranking

Pro Plus are those little caffeine pills that "help" you study and whatever other wild claims students believe.

Also causing sleepless night and palpitations, just like the impending AI bills.

[CrazyOldCatMan]

Re: The well-known name ranking

Pro Plus are those little caffeine pills that "help" you study and whatever other wild claims students believe

They don't work for me - and neither does chain-drinking doppio espressos :-(

(Yes - I have a weird metabolism. Caffeine doesn't keep me awake and never has done, even in the long-distant days where I had, on average, one coffee a month.)

[Jou (Mxyzptlk)]

Re: The well-known name ranking

> Yes - I have a weird metabolism. Caffeine doesn't keep me awake and never has done

You are not alone here. Never worked that way for me either, so I started coffee very late. I have reached the five, and I started ONE morning cup (~100 ml) of hot milk + one small spoon Caotina + 1.5 small spoon instant coffee to get my digestion system in gear. No change on the "awake" front though.

[Wade Burchette]

Re: The well-known name ranking

"It gave some pretty mediocre suggestions. Didn't really solve the problem, I still had to do most of the work myself."

Never would I have guess that Microslop made a slop AI that produced slop code. If only there was some recent history of Microsoft's coding quality to draw from ...

[BinkyTheMagicPaperclip]

Who could have predicted this would be a bait and switch?

Who could have thought this would follow the same rules as the gig economy, or online dating sites?

Well, apart from anyone not blinded by 'AI' that is.

If you must do it get an AMD Strix an Nvidia DGX Spark (or spend a ludicrous amount on a server) and host it locally. If the local models can't do what you want, well, all these cloud services can just keep raising the price until the pips squeak, can't they?

The question is whether the firms offering LLMs have attached enough bait and lock in to pull off the switch. Frankly, I don't think they have. I wouldn't bet money they've gathered enough buy in to avoid heavy losses either.

[Bebu sa Ware]

Re: Who could have predicted this would be a bait and switch?

[Github Crapilot] would follow the same rules as the gig economy, or online dating sites?

Painful and unpleasant fornication all round ? Not as though no one had been royally screwed by Microslop previously.

[BinkyTheMagicPaperclip]

Re: Who could have predicted this would be a bait and switch?

In the case of the gig economy, the workers get screwed, with not enough pay for too much work.

Online dating sites - oh, you'll *definitely* get screwed this month honest - but not by someone human. Maybe if you pay for Dating Pro Max 3 someone will notice you? Basically they deliberately keep you on the site, and de-emphasize searches unless you pay. For some sites unless you're an adonis, to see a better chance of being noticed you need to go beyond the 'let me message as many people as I want' tier to the 'your message will be prioritised' tier, plus the 'pay extra so the recipient can read your message even if they're not subscribed' tier because opposite sex dating, at least, is incredibly asymmetrical (not that many people help themselves, even when the dating site itself isn't trying to ruin things)

[martinusher]

Re: Who could have predicted this would be a bait and switch?

Its the same model that's used by every other 'service', including on-line gaming. (The trick with gaming is to set up the loss so it looks like 'almost a win'.)

[Elongated Muskrat]

Re: Who could have predicted this would be a bait and switch?

Sadly that asymmetry comes from the asymmetry that exists outside the world of online dating apps, which has been summed up before as men fear that the woman will reject them, while women fear that the man will kill them. To a lesser degree, men are far more likely to be perpetrators in abusive or manipulative relationships than women (although some women are by no means incapable of these things, and abuse can exist in all sorts of relationships, not just heterosexual romantic ones).

Dating apps do nothing to help address the underlying problems here, all they do is automate the process for profit.

[Boo Radley]

Re: Who could have predicted this would be a bait and switch?

Opposite sex dating sites? Are gay dating sites doing it differently? Why be specific unless you know something about gay sites... Just saying...

[Korev]

Re: Who could have predicted this would be a bait and switch?

I assume that the same sex dating sites screw[0] their customers in the same way the ones targeting straight people do. Maybe there are also extra things they do too.

I'm guessing (like me) that the OP is straight and therefore doesn't know much about the gay sites.

[0] Pun intended

[Elongated Muskrat]

Re: Who could have predicted this would be a bait and switch?

Well, you're hardly going to get gender asymmetry on sites where the users are matching with people of the same gender, although I would imagine there are other forms of stereotypical asymmetry at play here (top/bottom, butch/femme, etc.), it's just that they're not going to be so immediately and obviously derived from the generalised gender imbalance that comes from "the patriarchy".

[simonlb]

Re: Who could have predicted this would be a bait and switch?

Remember, those "heavy losses" will be paid for by someone everyone else, and they will be orders of magnitude worse than any losses from the 2008 crash. However, as with all the current late-stage capitalism, none of the glut of video, memory or storage devices that will flood the market afterwards will be reduced in price for 'reasons' so your next PC will still cost upwards of $3,000 even though it will be only mid-spec. How else will they claw back the insane amounts of money bet on this AI bubble?

[that one in the corner]

Re: Who could have predicted this would be a bait and switch?

> How else will they claw back the insane amounts of money bet on this AI bubble?

By demanding government bailouts, just like the last time? Yeah, it is still the same "everyone" who is footing the bill, but the marketing costs are lower for bungs than ads.

Especially as

> the glut of video, memory or storage devices that will flood the market afterwards

won't work in your $3,000 PC. Best plan to get yourself an empty 19" rack before the price on those skyrockets (or Ikea sells out of LACK tables).

[David Hicklin]

Re: Who could have predicted this would be a bait and switch?

> By demanding government bailouts, just like the last time?

Governments are also flat broke - no bailouts coming this time as they can't afford it

[retiredFool]

Re: Who could have predicted this would be a bait and switch?

Bailouts could happen I suppose. The new Fed Governor Warsh is going to be an interesting watch. On the one hand, he wants to lower rates. On the other hand he wants to reduce the Fed "portfolio" of debt. Problem is if you cut rates, and the fed won't buy the debt (ie print money) what happens when the UST can't sell a debt offering?

[MonkeyJuice]

Re: Who could have predicted this would be a bait and switch?

It should always have been clear that you are either doing most of the heavy lifting, or you were burning 6x what you were actually paying for and sometimes lucking out.

Long term the only sustainable solution would be local LLMs. The problem is these just don't work well enough yet by a long shot.

We simply don't have the hardware architectures available that do complex coding tasks with some probability of success that don't also slurp upsettingly large kWh that the majority of SMEs can afford*. Maybe the new GB300s will help, but I suspect they'll be north of $20k for a basic, slow, memory constrained box when they do finally ship, and since even the frontier models aren't really there yet, where would you even get the models to run on it?

* A friend at a rather large plc actually has on-prem Anthropic models running on enormous racks of GPUs- apparently last month the bean counters sent a memo out saying 'uh, can you try and reduce your token usage to about 20% please', so even the big players are baulking.

[Benegesserict Cumbersomberbatch]

Re: Who could have predicted this would be a bait and switch?

* Conclusion drawn from the bean-counters'memo: AI is <20% as effective as it was marketed / budgeted for. Who would have thought?

[Jonathon Green]

Re: Who could have predicted this would be a bait and switch?

“ If you must do it get an AMD Strix an Nvidia DGX Spark (or spend a ludicrous amount on a server) and host it locally. If the local models can't do what you want, well, all these cloud services can just keep raising the price until the pips squeak, can't they?‘

Hands up anyone who thinks that licensing for on-prem AI won’t (one way or another) go to a usage based model, like other on-prem services have…?

Yeah. Thought so… :-)

[AtomicDog]

Re: Who could have predicted this would be a bait and switch?

The "frontier" models may well end up charging on a per-usage basis even when running on-prem - but there will still be free models available that will make a decent enough job of it that they'll have to keep the prices vaguely reasonable if they go down that road.

[sal II]

Re: Who could have predicted this would be a bait and switch?

"The question is whether the firms offering LLMs have attached enough bait and lock in to pull off the switch."

They don't have choice, Antropic and OpenAI are about to go public, which will soon trigger the collapse. Once people see the real numbers and realize the emperor wears no clothes it's game over for the vast majority of these AI startups. They now have a limited window to extract as much as they can before the pop

[the GCHRD]

Your Skynet Protection account is overdrawn.

Skynet Customer Service "Yes, we're very sorry to hear you're having trouble adapting to the new Skynet Protection - Dynamic Pricing System. Your account has been suspended until your next payment clears.

You may be experiencing these problems, because you are in a high conflict zone and your protection is placing a higher demand on our systems. Please consider upgrading to our new Skynet Max account to ensure your protection is never interrupted again. (requires AAA credit rating or similar assets) You may also be able to reduce your billing by moving to a lower conflict zone or hiding under a durable surface, such as a rock."

the GCHRD "This is all coming much faster than I expected. Being barred from a service account is one thing, but I have grave concerns for democracy and society, unless or until effective global regulation is put in place. We need to wrest control from the delusional hands of the bros"

[Expect Great Things]

Copilot Turbo Platinum Select

Given that Copilot’s going to let each Master of the Vibes replace, what, 5?, 10? , 15? developers this seems like real value. What we’re going to need is a real top end Copilot plan that’s priced at, a still entirely reasonable, $50K/month.

[DrewPH]

Slop creators getting burned. This is delicious.

[Ryan D]

Predictable outcomes

My executives have, of course, gone overboard on AI. The CFO has been building agents to help with everything under the sun. Some of it might even be useful…

The same CFO who will nickel and dime on licensing, has just woken up to the reality (nothing like panicked email exchange first thing in the morning)that the bills are now going to start rolling in and look a lot like those nasty cloud bills.

They never learn.

[CrazyOldCatMan]

Re: Predictable outcomes

and look a lot like those nasty cloud bills

Previous orkplace went all-in on 'da cloud. Because it meant that they didn't need to look after on-site servers with the attendant costs (engineers, 'leccy, aircon [1], hardware replacement etc etc etc).

And, at the start, the costs were quite reasonable. Mainly because we'd only moved a small proportion of our systems and data to Azure. Then the costs (compute and, especially, storage) started going up and up - because they just replicated their on-prem setup in Azure, forgetting trivial facts like that you pay through the nose for *everything*.

By about a year before I left (end of last year) their monthly Azure cost was about what they had spent of hardware and support for one *year*. And still going up. Their annual Azure bill could have populated the server room with as much high-end kit as they could have conceivably needed along with all the people needed to keep it happy.

They were starting to migrate stuff back to on-prem as I was leaving in a bit to try to save money. I almost smiled..

[1] When they remembered to pay the maintenance for it. Or for it to be fixed at vast expense when it failed because they no longer had a maintenance contract..

[Daniel Pfeffer]

The chickens are coming home to roost...

... and everything the Cassandras said is turning oit to be true.

[Pascal Monett]

Re: The chickens are coming home to roost...

And it couldn't happen to a nicer bunch.

I am very happy to see that this is happening. AI needs to die and, if businesses are going to start looking at the bills and leaving, then the death of AI and the end of those gigantic, useless, water-guzzling factories will finally come to be.

And I can't wait for the day.

[The Central Scrutinizer]

And there we have it in a nutshell. You use AI, we screw you. LOL ad infinitum.

[Juha Meriluoto]

This "AI" crap is worse than drugs... All your money us gone, and you don't even get to hallucinate yourself...

[Anonymous Coward]

>This "AI" crap is worse than drugs... All your money us gone, and you don't even get to hallucinate yourself...

So, what you are saying is that using AI is like living with an addict?

[andy the pessimist]

just checking

Can i assume that regular (non AI) operations are cheap and the AI operations are expensive?

Can the person disable AI operations?

Do they publish the costs of operations?

[_wojtek]

bait and switch...

basically Mafia / drug dealer scheme... five out FIR free and then squeeze hard...

> "It gave some pretty mediocre suggestions. Didn't really solve the problem, I still had to do most of the work myself"

so basically doing twice the work (the work itself plus babysitting incompetent crap)? tell me about wasting time and resources...

[Jamie Jones]

Re: bait and switch...

Its always the same.. First vendors of software messed with formats to lock you into their office software.

Then cloud service did similar.

Now AI companies.

Basically, the same old gameplan, yet the management fall for it every time

[Paratiritis]

Obligatory

When working with the dotnet part of the Microsoft Agent Framework, which has been released to production, you constantly hit new cases of API changes, introducing breaks in existing code, which are not reflected in the official documentation or the samples. You get the feeling that this is intentional, a kind of dark pattern, where the use of AI assistants is becoming obligatory.

[ComputerSays_noAbsolutelyNo]

Re: Obligatory

Active

Programmer

Interference

[Andy 73]

Small question...

Am I allowed to find this funny?

[BartyFartsLast]

Re: Small question...

I think it's obligatory

[druck]

A rare moment of joy

Even with all the shit going on in the world, this really brought a smile to my face.

To all the AI slop slingers; welcome to reality, now go use your fucking brains.

[Anonymous Coward]

At least now we know ..

.. what Martin Shkreli has been doing after he was released from prison..

[BartyFartsLast]

Roosting chickens return home

Well, somebody has to pay back the trillions invested in this dead horse.

Might as well be the suckers who got addicted to it.

[Anonymous Coward]

Re: Roosting chickens return home

And this is a dead horse you can't even use for lasagna.

What, too soon?

:)

[seven of five]

Re: Roosting chickens return home

Hmmm, Lasagna...

[David 132]

Re: Roosting chickens return home

I hate Mondays.

[Kurgan]

Re: Roosting chickens return home

You don't put dead horse into a lasagna. Only dead cow and some dead pig. Italian here, I know how to make lasagna.

[David 132]

Re: Roosting chickens return home

...and if you do have to put horse in, and need to cover up the flavour, add soft cheese.

Specifically, Mascarpone.

(Mask-a-pony... geddit?)

Oh, tough crowd. Swing and a miss. I'll take my coat and go.

[Ken Shabby]

Re: Roosting chickens return home

Horse is too much quality, it needs Donkey

[David 132]

Re: Roosting chickens return home

Really? I always think that tastes like complete ass.

[CrazyOldCatMan]

Re: Roosting chickens return home

Horse is too much quality, it needs Donkey

Mortadella in lasagna? Are you mad?

[seven of five]

Re: Roosting chickens return home

> Italian here, I know how to make lasagna.

Oh, great. Another question, please: The Pineapple on the Pizza - slices or cubes?

g,dnr(unlikehell)

[CrazyOldCatMan]

Re: Roosting chickens return home

The Pineapple on the Pizza - slices or cubes?

Sliced finely and seared in the pan before putting on the pizza is best..

(Yes, I do like pineapple on pizza - but if you put it on straight from the tin (or cut from the fresh pineapple) it will make the pizza soggy..)

[Anonymous Coward]

Re: Roosting chickens return home

Easy to fix - leave them in the tin.

:)

[CrazyOldCatMan]

Re: Roosting chickens return home

You don't put dead horse into a lasagna. Only dead cow and some dead pig. Italian here, I know how to make lasagna

I'm not really a fan of lasagna - something about the texture of the lasagna sheets in the middle makes me want to hurl.

A rich ragu [1] now - served over well-cooked pasta [2] - a thing of sublime joy. Especially when accompanied by a nice glass [3] of good red wine.

[1] Using a mix of pork and beef mince naturally. Or Italian fennel sausage.. I'm happy to cook either. And no, I don't use shop-bought sauces - why would I when it's so easy to make? Not a process to hurry though.

[2] Preferably fresh egg pasta (and no - life is too short for me to make it when the shops have perfectly good stuff) but dried pasta is also acceptable. And keeps for a *long* time.

[3] A bottle is made of glass - does that count?

[midnitet0ker]

Muh Subsidized Workflows...

I was wondering how tech CEOs were going to make the maths math on profitability vis-a-vis revenue and it turns out *they're not*. Once people start paying what it costs to run these models, AI is going to prove prohibitively expensive for nearly all clients except for perhaps the U.S. military with its bottomless budget. Regular Joes are going to bolt for the hills as evidenced, finding solace no place as the cost-benefit paradigm catches up to them (especially the management class). It would have been cheaper to hire a code monkey straight out of college and that code monkey would probably have the sense not to plow through your inbox deleting everything in sight or wiping a production database in 6 seconds flat (as I've read of AI agents doing).

The AI that CEOs and management forced down their employees' throats is going to prove ruinously expensive compared to that recently graduated kid they laid off in favor of Claude a while back. I foresee management come crawling back to rehire laid off employees and hopefully wages spiking when jilted former hires tell their erstwhile bosses to get bent, creating a wage spike anywhere businesses went in head-first on AI and ended toes-up on the bill.

[CrazyOldCatMan]

Re: Muh Subsidized Workflows...

forced down their employees' throats is going to prove ruinously expensive compared

Especially if the employees do a bit of creative malicious compliance to the AI-first rule and make sure to tokenmaxx at every possible opportunity [1] in order to drive up costs..

"Well, you told us to do it! I was just obeying orders!"

[1] Claude, please tell me what the weather is outside at the moment, integrating all available weather sites to ensure a plurality of viewpoints..

[ForthIsNotDead]

Similar experience for me

I've found it very useful. But it's now just too expensive, so I'll go back to the old way.

However, code prompts in co-pilot remain free (at least, for now) so you get the benefit of intellisense on steroids, even if you've blown through your allowance. Sometimes its very useful, other times not. It's excellent when you're re-factoring code. If you use descriptive function names it will often write an entire function for you perfectly.

But yeah... I think the days of giving it a prompt and sitting back and watching the magic happen are over for me. It's just too expensive. I have an annual subscription which I won't be renewing.

[Bryan W]

Re: Similar experience for me

Not really. Rather have IntelliSense et. al. stay in its lane and suggest identifier names from a list of actual identfiers in scope and not try to fail at doing my f'n job while I'm doing it.

[Anonymous Coward]

This is an issue....

People have been able to do things they were not able to do without Ai. Coding in particular. Sure the code may not be great, but it works and is not something many people would never have tried/been able to do with out Ai.

Now all those people on a $10 Copilot subscription need to go to something else or stop doing what they did. Expectations have been set and situations created which they can not live up to unless they go for a potentially much more expensive subscription.

Looking at the programmers I know who use Ai. When I see how they use it, its very much an accelerator for them, saves a lot of typing and boring stuff. They could still do their job with less or no Ai.

[Crypto Monad]

And therefore, if it saves them even 1 hour per month, it's got to be worth at least $100 per month - that I guess is the selling point. People buy multiple monitors or fancy keyboards to make themselves even a tinier bit more productive and/or comfortable in their work.

Personally, I'm more comfortable with an editor that just accepts my keystrokes verbatim.

[CrapPilot]

Except that it's not $100/mo. I was over $450 before the end of the 3rd. It is not a small price increase - it's orders of magnitude off.

On the plus side, Amex is friendly. Opened a dispute for the upgrade to the max plan (that *alone* is 100), last month's usage...which was like 150 - total, and will do a new one for when this one gets charged.

...and switched back to my Claude Max 100 plan.

[An_Old_Dog]

AI Code Quality

Sure the code may not be great, but it works

In these days of exploits, ransomware, and malware, low-quality code which does work, is not good enough, unless your company doesn't care about being pwned.

[Elongated Muskrat]

Sure the code may not be great, but it works

This. This here is the problem.

AI generates naïve implementations, or inefficient code, or code that works in the specific case but not the general one.

To write functional, maintainable code, you have to understand the requirements the code is implementing. That means not just the golden-path "it works" case, but handling error conditions, exploits, and bad actors, as well as considering performance, where it is appropriate to do so (and yes, not considering it when it isn't, Donald Knuth).

A fancy pattern-matching algorithm doesn't do any of that, it generates code based on a set of training data, that passes its immediate requirements, without employing any thought whatsoever, and until AI actually has genuine intelligence (which is arguably impossible in a finite state machine), it won't be able to do anything other than appear to be doing this on a superficial level.

[JimmyPage]

Very little sympathy, I am afraid

downvote all you like. As you explain to me what it was about the previous 30 years of MS behaviour that made you think "You know what ? This time thins will be different".

I have been using "AI" for the past 5 years in a professional capacity. And my original observation that it's most useful feature is to work around the shite that is modern IT still stands.

[Jonathon Green]

Re: Very little sympathy, I am afraid

So The Song Remains the Same… :-)

[Bryan W]

Grab your pitchforks

This disconnect on Eh Aye between idiots playing at being coders because STEM and "those who can" here has been echoed since the first AI code assistants promised vibe coders they didn't have to RTFM.

These wannabes have been spending this time sitting at our tables, in our meetings, in our forums, offering their (read: Claude's) "expertise," burning through our budgets and acting like they know what they are doing by making engineering decisions with us when all they are is a mouthpiece for whatever LLM service they are addicted to. They won't answer questions in real time because they need to ask their outsourced brains off to the side and will try to act like they came up with it. They are absolutely worthless as problem solvers but mediocre at vast quantities of mindless CRUD work.They are either too inexperienced or too stupid to identify these two separate yet interdependent facets of the work and so think this is an "improvement." It is shortsighted and we've been saying this the whole time.

I can ask the LLM and get the same answers they'll give me, errors and lack of insight included. You want to replace humans? Find all the lazy non-thinkers who glommed onto and pushed for this lazy crap. There's your cost center to cut right there.

Every decent engineer I've worked with has only found it useful for replacing that whole "Google for answers" part of our jobs. This fact has not changed no matter how hard they've been hyping the new drug's "agentic" capabilities.

No one worth a damn is using these integrated agents on their actual code base. I copy/paste from chat occasionally and almost every single time it is wrong and needs fixing in some deviously subtle way. Who TF actually getting a job done wants to deal with that? The idea of letting this thing go off leash on any code my name is associated with is ludicrous and I say this as someone who rates their own code as an "ancient dumpster fire that somehow works pretty well."

Finally, and call this a personal point: Anyone who willingly outsources their critical thinking like this doesn't deserve the title engineer or to sit at the same table. These are the people who sit back and let everyone else do the work. Paycheck collectors. Bumps on the log. No wonder the idea of some computer thingies doing their job for them sounds so enticing.

And you C Suites want to pay for this? LMAO

[Anonymous Coward]

Re: someone who rates their own code as an "ancient dumpster fire that somehow works pretty well."

... Are you one of the other voices in my head?

[CrazyOldCatMan]

Re: Grab your pitchforks

They are absolutely worthless as problem solvers but mediocre at vast quantities of mindless CRUD work

It reminds me of a lot of the y2k work that I saw around me when I was doing it - people who could take a script written by someone like me [1] and do the actions and record the responses.

They generally didn't have a clue about the context of what they were doing or, sometimes, what the responses meant but they had their uses.

Then the y2k stuff ended and my available IT contract rates dropped from ~£40/hour down to ~£15 - because all the morons that thought that two years of sitting in front of a computer doing menial stuff gave them the ability to sell their 'skills' on the contract market. Without a clue what even their minimal skills were worth. Thus killing the market for the rest of us, even though of us with lots of good skills in a wide set of technologies.

So I turned my back on the whole EKS thing and went back to being permie. OK - it still wasn't £40/hour but at least I didn't have to worry about all the admin and hassle that went with being a contractor. And, when IR35 came along, I wasn't affected.

[vtcodger]

View from a distance

Being retired, I thankfully don't have to worry much about AI. But I do question how any well run business (if any such exist) can possibly embrace a technology whose actual costs are substantial, unknown, and apparently unknowable. Makes one think Ed Zitron is quite likely correct https://www.wheresyoured.at/the-revenge-of-the-business-idiot/ when he tells us that we are being ruled by f*cking idiots.

And actually I should think that part of the problem goes beyond AI to Software as a service in general. If your business workflow depends on someone else's computer and code, you are at their mercy. They can hike your prices with little or no notice. They can drop capabilities you need. Again with little or no notice. If they fail, your business may well be collateral damage.

My feeling would be, if/when you can bring AI in house, evaluate it it and, if it makes sense for you, embrace it. Until then, avoid it. It's potentially and maybe actually toxic.

[MonkeyJuice]

Re: View from a distance

I think the part that pisses me off the most is the veiled threats. "You either embrace this or go the way of the weavers and lose your job!111". This is utter bs- because the tech is just nowhere near ready yet. It's patently clear that human factors (the stupefaction that sets in after a few hours of just letting the machine do its thing, or the fatigue of hand holding it), plus the inability of LLMs to coherently keep the big picture in mind are major blockers for it to justify the level of dependency some people are getting themselves into, before we even touch on the vendor lock-in and price spiking.

For those still not on the bandwagon but feeling pangs of FOMO, I'll say this- you can save your sanity and wallet for now. This stuff is a decade off from being the kind of game changer decent refactoring IDE tools or git were. When/if it is ready and worth a look, you will know, and the hype will have died down fully. It's a curiosity, but it's definitely not a coherent timesaver yet, and it's vastly more important to be familiar with your codebase, which AI can dizzyingly replace or bloat at speed. We appear to be missing a few fundamental parts before there is a cost-benefit that is comfortably a no-brainer. Good version control and refactoring tools never needed a hype cycle and such aggressive marketing- people adopted those because like all good technologies they obviously and unambiguously improved quality of life.

[Kurgan]

Re: View from a distance

The issue is that AI will destroy enough wealth that your pension will evaporate, too.

And about the If your business workflow depends on someone else's computer and code, you are at their mercy part, well, this is how every business (not mine) works. Even before AI, we have become cloud-dependent, and subscription-dependent. At the flick of a switch from MS, Google and AWS, 90% of the businesses, schools, and governments of the western world will disappear in a puff of smoke.

[Nate Amsden]

Re: View from a distance

Same thing happened to public cloud. Just at faster rate with LLMs. I first wrote about it in 2010 then wrote more extensively about it late last year on my new website https://cultofthe.cloud/ . Dozen articles, 40,000+ words, and I've not used an LLM for anything on that site or anywhere else in my life yet anyway. Also no ads, no trackers of any kind. I haven't even looked at how many hits in my web logs for the site. I post links to the site on LinkedIn in comments on an almost daily basis since last year.

At a big scale take the Geico insurance company, after spending a decade trying to make public cloud work they started reversing course last year I think after balking at a $300 million cloud bill. Story of them and others in one of the articles on the site. Similar with SAP though it took them 5 years to realize. Problems exist on small scale too, one of the articles is about my first hand experience at multiple companies over about 2 and a half years. Moved out of public cloud 14 years ago...i even host the website on my own Dell hardware in a colo I pay for.

I made the site because there's far too much info to convey in any comment box. I have written many such comments here on el reg over the years but often people have other questions too.. so I put everything in one place.

[Pirate Peter]

as I have said before AI pushers use the same MO as drug pushers

they give away free or very cheap loss leaders to get you hooked

then once you or your company are dependent on the drug / AI they ramp the price up as they know there is little if anything as you / your company is addicted and all you can do is cave in to their demands / extortion

and getting clean again will involve trying to work out and remove a load of the slop that was created while you / your company was addicted

same as smoking and social networking I've seen the crap it causes and never partaken (despite company insistence) in AI slop, after doing various exams and seeing behind the curtain there's no intelligence just complex predictions to work out B follows A in the alphabet base on what its tables have for the probability

[Captain Slog]

Re: as I have said before AI pushers use the same MO as drug pushers

I said God damn, God damn The Pusher man

[JWLong]

Re: as I have said before AI pushers use the same MO as drug pushers

>>I said God damn, God damn The Pusher man<<

https://www.youtube.com/watch?v=Zv6PY1BQLBE

One of my most favorite songs in high school.

[Pete Sdev]

ha ha ha ha ha ha ha ha

Oh my aching sides.

That the current costs of providing "AI" is probably several times revenue until now (see Ed Zitron's analysis) should make large price rises no surprise whatsoever.

For a GoT take:

"You vibe-coded this?"

"Lots of people vibe-code"

"Lots of cunts"

[Anonymous Coward]

Idiotic managers - but I repeat myself - hurtled right past LLMs as coding aid, a (potentially) useful, still almost reasonable tool, all the way to the Dunning-Kruger fantasies of autonomic AI "teams" that are pretty much designed to waste effort to try to reproduce the existing team setups and produce nothing of real value beyond what the human in the loop is capable of processing - inefficiently, because of the ridiculous business consultant-designed models of how AI software development shall look like. Meanwhile the new versions of those commercial models can't seem to do anything but throw more processing and internal checking at their faults, with each version significantly slower and more expensive to run than the previous one while making the same kinds of category errors that always plagued the stuff.

Watching this play out is like watching a relative get ensnared in some MLM scam or become a born again/cultist. I have lost most all the respect I used to hold for so many people the past few months.

[steviebuk]

bait and switch

“This is a staggering shift from a 'predictable subscription' to a 'stressful meter-based' service that hinders my productivity rather than helping it,”

What did you expect. Its what they always do. Pull you in with either free or other such service. Then once enough of you are trapped and can't pull out easy, they change the model so you have to pay loads.

Arseholes. Can't wait for the bubble to burst.

[Anonymous Coward]

Re: bait and switch

And this one was even more predictable than in most situations of this type, because anyone with the most basic knowledge of LLMs would be aware that these subscriptions weren't covering the actual costs, never mind making a profit! It should therefore have been obvious to everyone what was going to happen once they had got enough people hooked.

[BasicReality]

Nice to see good news like this, especially when the idiots at my company are making a push for us to start using copilot.

[Sherrie Ludwig]

Nobody here lived through the 1960s drug era

Every smack dealer ever: "The first one's free."

[An_Old_Dog]

Re: Nobody here lived through the 1960s drug era

Mother: "If Bobby jumped off the top of a fifty-story building, would you do that, too?"

Son: "...... Nooooo."

Drug addicts made an initial choice.

So did the fad-followers and greed-heads who wanted the "something for nothing" promoters suggested users would get from "A.I."

[frankyunderwood123]

Corporate is gonna corporate

"Sorry, we can't afford any more <insert-service-here> licenses - budgets are tight!"

"We're going to migrate everything off <insert-service-here> to cut costs this year - budgets are tight!"

"You'll just have to hold onto that old laptop of yours for another few years, there's no IT hardware budget this year!"

....

"Hey, we want you to extensively use AI to do all of your work via agents. Here, have a Copilot license and Claude Code one too."

"We expect you will be 10x more productive!"

"Use AI to review the code - but with human oversight, of course."

....

"Er, excuse me <team-mate>, your PR has 123 new files, 55 deleted files and 323 changed files in one commit - can you please break that down for me? Ah, whatever, I'll get AI to review it"

....

"We've noticed our bills are sky high and the quality of software has slumped, we're going to have to lay some of you developers off. Sorry, we'll try and reassign whoever we can whilst we restructure your department."

[Anonymous Coward]

You mean all those bit barns and electricity are not free? Who knew?

[WolfFan]

Over on Redit

There are a LOT of subs infested by AI astroturfers shilling for various AI 'agents'. The most common seem to be the anti-TurnItIn bots, trying desperately to convince gullible students that this bot will write all their papers for them, cheaply or for free, and the AI detectors will never spot them, and if by some chance they are spotted, this is how to convince the professor to give you the 'A' anyway.

My current favorite is the extremely obvious bot posting about how it has lost a $45,000 scholarship to a 'private university' because it had a paper flagged as 98% AI when it wrote that paper manually, and how AI detectors are the worst thing ever and it will be suing the university. And, oh, it had a full version history showing that it wrote the paper manually over the course of several weeks but the instructor refused to look at it because the instructor believed his AI detector. There are multiple replies to this lovely bit of AI-generated fiction, all (except for two other bits of AI-generated fiction) saying that this never happened and that having a version history is EXACTLY the first thing that professors look for when a paper gets flagged as being AI, and that a relevant version history is a good indicator that the AI detector has a problem. Many of the replies are from professors who state that when they get a paper which is flagged for significant AI writing (over 50%) the first thing they ask for is a version history... and, amazingly enough the students NEVER provide one. This would be A Clue. Many of the professors state that their universities explicitly request that students turn on version histories and that indeed the universities have pages on their sites showing how to turn on version histories in common word processors, with MS Word, Google Docs, LibreOffice, and Apple Pages being specifically mentioned as apps covered by the instructions. They also asked that the OP produce the name of the university in question. (After all, if the story was true, there's already a lawsuit out there and the OP has nothing to lose by naming and shaming...) Amazingly the OP has vanished softly and silently into the aether. No names, no comments, no nothing. Gee. I wonder why this might be.

AI writing is wildly incompetent and easily differentiated from actual real persons writing. AI is not, at this time, fit for purpose... except to detect AI. It's good at that. And maybe to spot bugs in large code bases. It seems to be good at that, too. Writing good code, not so much. Writing just about anything else, very bad. Those who push AI refuse to believe just how bad their products are, and try again and again without achieving anything except much merriment.

[a_yank_lurker]

MicroSlop going to MicroSlop

So one of the most vicious and predatory (criminal) organizations still acts like the criminal organization they are, I shocked I tell you, Shocked I tell you like Captain Louis Renault in Casablanca.

[Tam Lin]

My excuse is that there wasn't really anyone around to warn me about Microsoft in 1974

What excuses are the kids using these days?

[ecofeco]

Re: My excuse is that there wasn't really anyone around to warn me about Microsoft in 1974

You're holding it wrong?

[_olli]

Re: My excuse is that there wasn't really anyone around to warn me about Microsoft in 1974

Kids nowadays don't even know what is software. They just install stuff from the apple/play store to their smartphones and don't spend a thought on how it all works or who collects the money.

[Boris the Cockroach]

Wont be long

until windows itself is free but you'll have to pay for the extras....

Ai.... I can handle it.... i can quit ... anytime.... anytime.. anytime I want. just got a bit of cold right now

[Paul Hovnanian]

$6 ... Really?

One GitHub Copilot developer requested a single change to their project and burned more than $6, they wrote.

What do you think making "just one change" costs in meatspace? A pull request, your time to putter around with the code, maybe a peer review. And if you are a fully ISO- compliant shop, some testing. And documentation. The actual code part is cheap.

The problem seems to be: This is actually an accountable item now. The Boss will see it. And perhaps ask quesions. Can't use the "compiling" excuse any more.

[MarkMac]

Re: $6 ... Really?

It was always accountable.

The issue is that the seller shifted from a per-month fee to a per-transaction fee, with no choice given to the consumer. There's only one reason they did that, and its to increase revenue.

As for the time taken to make a single change to a project - all the stuff mentioned, peer reviews, ISO compliance, testing, documentation - still has to be done even if AI slaps the code together.

[spireite]

AI is like getting married

You start with something/someone new....

Each token/date is cheap

You subscribe/get married

with each upgrade/anniversary. you get less for your money

Eventually, you realise that the life subscription has decimated your account.

Proving....

IT IS THE SAME THING!

[MarkMac]

"Copilot max" - how about Copilot Zero, for those of us who want NOTHING TO DO with this slop.

[M.V. Lipvig]

Sweet Clyde - laugh derisively in their faces.

HAA, ha HAAAA, HA Ha haaa.

The only thing surprising about this is how quickly M$ did this. I would have expected this around this time next year. The early adopters were in, but the rest of the companies were just starting to dip a toe into the water. This is only going to hit the early adopters while the more cautious will abandon it.

That said, my company is pushing ai hard and they use copilot. Methinks it's time to get on board, right up to and including asking copilot when the best time of day to pick my nose is, and whether or not I should eat it.

[ecofeco]

Why did they wait so long?

Oh NOW they've figured out M$ is screwing them.

They is be geniuses!

[CrapPilot]

Re: Why did they wait so long?

The reality is they aren't screwing anyone. They are beginning ti show the ACTUAL costs of running this dumpster fire. It will get worse.

Anthropic pulled back dropping Claude Code from Pro+ over backlash. They are chomping at the bit to recover costs.

When you see estimates of $5k/mo in compute to service a typical $39/mo subscription - those economics will never work out

[Noodle]

I struggle to have much sympathy for people who were using a service clearly priced in a way that bore no relationship to the actual cost of said service and are now SHOCKED that the giant corporations behind those services are starting to charge them more realistic prices.

[Covid-20/21]

Claude 4.8

Sorry, I hardly have any sympathy for someone who says “ just needed Claude 4.8 to help fix a couple things on a site I'm editing” and then wonders why their token allowance has rapidly diminished. The model that’s 30x in cost. Don’t use 4.8 on a ‘site’, use 4.6 it’s really capable and 3x, tokens last longer. This is why MS is charging by the token, the stick to make people aware of what model they are using ( and actually make money ). Use tokens carefully and they will last.

[Rogerborg 2.0]

Re: Claude 4.8

"Useless answers are affordable."

[Rogerborg 2.0]

MegaCorp Inc., last month: "Use AI more, it's cheaper than you, monkey!"

MegaCorp Inc., today month: "Stop using AI, monkey, we can't afford it!"

I'd already blown through my monthly tokens by the time I found out.

[[e-mail]]

Don't you hate it when Microsoft has a point.

At some point in time it was inevitable that end users would need to pay the entire cost of AI and a bit more for profit. That's the way the world works fellas. It you don't believe me, ask ChatGPT.

[herrjdev]

Honestly devs need more visibility on their credit usage. That’s why I started working on an extension to track my usage and show me useful stats like daily/weekly usage. So far it’s been helpful. It’s called Tokenyst on the VS marketplace if you wanna try it yourself

[herrjdev]

Usage stats

Honestly devs need more visibility on their credit usage. That’s why I started working on an extension to track my usage and show me useful stats like daily/weekly usage. So far it’s been helpful. It’s called Tokenyst on the VS marketplace if you wanna try it yourself

[Anonymous Coward]

I hit $355 in usage on day two. Day one was initially showing 65, then ballooned to 80 something, so I upgraded to max.

I then proceeded to tell crappilot to make X look exactly like Y, then to even just USE component Y and add a new property to it - over, and over,, and over as it insisted they look exactly the same. I assure you, they did not.

What did look the same was my ever increasingly annoyed chats and the language you'd hear during Fleet Week or at around 2AM at a dive bar where you're more likely to he shived than to get something resembling a cold pint of beer.

[Scene it all]

Who asked for this?

I don't use Github any more but were people banging on the doors ASKING MSFT for an AI assistant?

[Slions]

Good news, bad news

To the vibe coder crowd, it's probably for the best. You can slow down a bit and try understand some of the code your agent has been spitting out. Don't worry, nobody will miss your low effort: "I build a tool that helps you build your tooling tool, curious what you think about it" posts.

To the higher than thou, old school, better than most wise crowd, some of whom have not even tried AI or apparently can't understand how to boost their own productivity with a coding agent despite their infinite wisdom, I have bad news for you: coding agents are not going anywhere, they are here to stay, one way or another, look for a local LLM near you.

[cookiecutter]

why is anyone shocked?!

these fools have spent $700 thousand million this year on capex alone and want to spend $1000 thousand million next year. based on shit harry potter fan fiction for a product that at best can tell people how to look themselves and offer to write the suicide note for them.

what did people THINK would happen?!

"oh we're going to carry on letting you cost us $13 for every $1 we make?"

I'm not upset at the AI Grift (i am); i'm just genuinely shocked that i'm not shocked that developers (a bunch of people who are surprised when you explain DNS to them & shouldn't be allowed near anything more complicated than an etch-a-sketch) are surprised that suddenly MS, OpenAI & anthropic are fucking them over now on cost.

i've been going on about how AI is a giant grift for 2 years now. i'm genuinely interested to find out who the company that spent $500 million on a month on anthropic are because i'm going to genuinely piss myself laughing at them but also the idea that business idiots are already jumping onto the next grift "quantum" & knowing that MBA fuck nuggets will play the exact same playbook to get other MBA fuck nugget business idiots to spend $millions on something that doesn't work

[JLV]

A lot of AI issues…

… will go away once it’s properly priced.

Less AI slop content. Less malware. Less vibe coding BS. Less Doctor Jesuses (I wish). Less AI tai chi ads (I wish). Less layoffs when folks consider the alternative of getting a $500M bill :-(

What’s uncertain is whether there’ll be a hard landing for investors as people realize just how expensive it is.

I suspect many will look into running models locally via Ollama or the like, putting more pressure on hopes for golden geese.

[Rogerborg 2.0]

Re: A lot of AI issues…

Having just burned through my monthly Copilot budget in 2 days, I am indeed now testing models in Ollama. Clanker-slave to do dumb-work while I do the brainwork is just another useful tool. Catch up, or be left behind.

[imanidiot]

Welcome to the real* cost of "AI"

It's going to take a while but most other "AI" are likely going to follow suit and costs will rise precipitously. And then the majority of use will stop.

*Note, not actually the real cost yet. It's going to get much more expensive.

[spireite]

Re: Welcome to the real* cost of "AI"

Plenty of pips left to squeak....