Be a hater all you want, AI's here to stay

AI gets things wrong, it kills jobs, and when the bubble bursts, it will take the economy down with it. When all's said and done, however, AI will still be standing and fundamentally change our lives and work. Get used to it. I get it. There's a lot to hate about AI. Personally, I'm both an AI user and an AI hater. Yet at the …

[PinchOfSalt]

I'm not sure it's fair to say that people are increasingly wanting to be told a comfortable lie rather than the truth.

This is normal. Indeed, it is almost how we remain sane. The world is a dangerous place, so numbing that to one degree or another is a survival tactic.

What's probably more challenging is that the 'world' is further away than ever. The bits we don't like are place geographically a long way from us so we don't need to face them each day. Water treatment plants, power stations, overhead lines, abattoirs, slave labour production facilities are all kept away from our eyes and ears. Yet, we happily take the benefits and treat them as rights.

This leads to a very unrealistic perception of the world, and hence comfortable lies become easier to spread since we have such a poor grasp of reality.

[Brewster's Angle Grinder]

"...the 'world' is further away than ever..."

Isn't it the reverse?

In ye-olde days, the bits of the world that could effect you were your immediate neighborhood. The further you moved away, the less effect than would have (modulo occasional exceptions like massive volcanic eruptions). Sure, you might have famines but you had a kind of control.

Now, we in the West are rich enough to avoid famines etc... but local events on the other side of the planet might have an effect on our daily lives. Our neighborhood is the whole world. It takes a helluva lot of bandwidth to manage that. And the complexity exceeds the simple stories we, as a species, used with while the world was small.

[steelpillow]

Nearer in one dimension, further in another.

"Look, Dougal, I'll try one more time. This cow is small, that cow is further away."

[StewartWhite]

Down with this sort of thing!

[Anonymous Coward]

This leads to a very unrealistic perception of the world ...

Yes, it does.

But ...

Do we really know who Perplexity actually belongs to or is controlled by?

eg: Alphabet Inc.

Asking for a friend ...

.

[Pascal Monett]

Re: who Perplexity actually belongs to

There is, unfortunately, a statistically insignificant portion of humans with access to Internet who ever ask themselves that kind of question.

Which is another part of the problem that we pull the wool over our own eyes and have a great tendency to accept having someone else do so.

[You aint sin me, roit]

Their CEO is a very nice man: "Srinivas said the incoming AI-related layoffs were a "glorious future" to "look forward" to, as it freed people from jobs they didn't like and gave them opportunities to pursue entrepreneurship"

Free at last! Now I can give up my well paid job and become "an entrepreneur"...

[Evil Auditor]

I think at least part the problem is that life, while it has been getting easier (not simpler!), it has also increased complexity: more choices, more decisions to take, and being fed an enormous amount of (mis-)information... which might be rather overwhelming and leading to a desire of being told a simple, comfortable lie rather than dealing with the complex reality.

_{p.s. It's only a short walk from here to voters' behaviours and the rise of certain populist parties, which do provide simple "solutions"...}

[Anonymous Coward]

life, while it has been getting easier (not simpler!)

Easier? Ask the student who has just left university/college and is not able to get a job. Ask those people who have to work as zero-hours packers or delivery drivers. Ask the families where both parents are forced to work full time, and still can't afford to pay for the roof over their heads.

There used to be an implied social contract: you got offered work, you applied yourself vigorously to that work, and in exchange you could expect a reasonable standard of living and some time outside of work to relax and enjoy life. This seems to be disintegrating.

And yes: since the mainstream political parties have failed to address this, the alternative/extremist parties are increasing in popularity.

[doublelayer]

Easier does not mean easy. You are often overestimating how many people got that social contract and what actually was included. Many of the changes in recent history haven't been about what low-wage workers experience but about the availability of high-wage jobs, especially relative to the experience needed to obtain them. Those who lost those experienced the most severe decline, but if you let that lead you to believe that that was available to everyone, you will end up having a much more positive idea of history than actually existed, and that will make it even harder to try to reimplement that today.

We already have a bunch of factors limiting our ability to put the society we want in place. As much as Trump is giving it a try, we can't prevent global manufacturing from being cheaper and people wanting things at low prices, which means a bunch of relatively high-paid manufacturing work isn't going to suddenly appear. If you recognize something as hard, you can put the increased effort into it that might let you accomplish it. If you convince yourself that it was easy before based on an overestimation of how much it existed, you'll treat it as easier than it is, and politicians like when you think that problems are easy because maybe voting for them is all you need to do.

[ChoHag]

> we can't prevent global manufacturing from being cheaper

Oh we can, but to do that we're going to have to accept that it means enriching the people who live in those cheaper climes rather than exploiting their desperation.

[doublelayer]

That's going to happen anyway, but there are a lot of people who earn and will accept, some happily and some less so, amounts that would be a pittance to us. Until they're all enriched to the same level we are, almost everything involving a lot of physical labor is going to be cheaper when they do it, even if the they switches countries a few times. And, if robots keep improving, that process is going to get dragged out. Similar things will happen with service workers and has already started with things like software which are easy to distribute. Denying it doesn't work, and fighting against it mostly doesn't work either, especially if those doing the fighting aren't careful about how they try it.

[Missing Semicolon]

" it means enriching the people who live in those cheaper climes". In your dreams. It just means enriching the interposed globocorps that get rich arbitraging low-wage economies with high-spending ones.

The only fix is to cut off their oxygen supply by taxing imports to make such practices uneconomic.

[BoldMan]

So you want to tax the consumers to harm the producers? Is tour real name Trump?

[Dr Dan Holdsworth]

What you also get when you destroy that social contract is somewhere like Russia. Over there people are so despondent about the future that they are simply not having children out of compassion for these unborn people; after all if life is utterly miserable and doesn't look likely to get better why create more people to experience this misery?

AI will, I think, make a few things easier but the bulk of the LLM applications will be too expensive for it to be worth while. Then you may see an increasing trend of people boycotting companies that use AI extensively, not least because spending a few hours arguing with an effectively brainless AI tends to get so soul-destroying that people quit using that company simply to save their sanity.

[SundogUK]

Speak for yourself.

[Anonymous Coward]

He did and you came back with an inane reply. Like almost every right wing muppets do.

[EricB123]

I live in "that part of the world", where people work 6 days a week to earn $300 a month. I'm retired, and even my small pension is larger than that. Quite the learning experience.

[andy gibson]

"The bits we don't like are place geographically a long way from us"

"The bits we don't like are place geographically a long way from us"

Not sure where you're based but that's not a thing in the UK. New builds are besides bypasses, railway lines, huge warehouses, sewage works all the time now.

https://www.bbc.co.uk/news/articles/c5ywdw800nzo

[Cubbie Roo]

Hate is a strong word, even though it's hollowed out my profession and forced me to switch disciplines. I simply don't want 'machine learning' in my life so have re-structured my compute accordingly.

And when I look at the folk in my immediate circle using these tools daily, and the noticeable effects on their mental acuity, I'm confident that opting out was the right decision.

[werdsmith]

I've got a relative in software development, up against a deadline and working through a moving target bug list. He is having to fully use his Claude Code because the shop just doesn't have the time to do it all with human coding. He is concerned about his skills, but that will have to wait until they move out of delivery and into maintenance.

[Elongated Muskrat]

I hope you enjoy maintenance programming, because that's going to be the main thing "AI" is generating for you. Your approach is akin to "ship the code as quickly as we can by getting an army of juniors to write it," and that's a fairly accurate representation of it, because my experience of AI-written code is that it will be replete with na­­­ïve implementations.

[werdsmith]

Well I don’t do maintenance programming, I’m not even involved in the project so you don’t need to hope anything. Just read better.

But I’m told, and I trust him well, that Claude is doing a fine job of quickly tracking down the reasons for the bugs in the human code and assisting the humans in making the corrections.

Best not to scoff and sneer because you don’t know.

[Throg]

Except we do know, because many of us have been here time and time again.

What you're describing is a rush to "code complete". Whenever that's used as a benchmark for delivery, the result is often a very long tail of test / fix cycles. If you're lucky, you'll have test and staging environments and a good team of testers. Sadly most of these projects also tend to cut corners there as well, so each cycle incorporates a deployment to production and end users as testers.

Every time you rush one of those cycles, code quality becomes worse and the tail becomes longer, until the code base turns into an unmaintainable ball of spaghetti.

These projects end up being death marches, but like a frog in boiling water the poor team don't realise it until months or even years down the road. The OP would do well to read Ed Yourdon's book of the same name. I think it's out of print but there are copies floating around the interwebs.

[matjaggard]

That's not really true. You can use AI to make a quick and dirty solution, sure. But you can also use it to make reliable and well designed software and to maintain that software.

For me personally the primary things it's done are make me able it apply my years of experience in more areas - for example I rarely care about if I know a specific language anymore - and to speed up delivery.

Don't get me wrong, it has downsides - whilst I still get the satisfaction of delivering good value for customers, I do miss the act of coding - I also worry about the next senior developers as the article mentioned, but overall it's a net positive to me.

[Elongated Muskrat]

you can also use it to make reliable and well designed software and to maintain that software.

You're really going to need to provide a citation for this, because my experience (and understanding of what it is and how it works) is that it can't do design, well or otherwise, only being able to regurgitate patterns it has been trained on, which, if you're lucky might fit your problem domain with a reasonable approximation.

If you think it can maintain software, you have clearly never asked it to provide a solution to a non-trivial problem and then fix its own bugs. This isn't a negative-feedback loop that reduces the errors on each pass; it is a positive feedback loop that increases them. This is down to a genuine physical constraint: entropy. Entropy doesn't care how fancy your LLM is, in this house we obey the second law of thermodynamics.

To remove errors from code requires an understanding of the error. Is it something as simple as a one-character typo (and these can be catastrophic, just ask the person who got a comma and a semicolon mixed up in the control software for the Polaris AX-1 test firing), in which case, it might be trivial to fix, or is it down to a misinterpretation of the domain requirements, which might necessitate an entire rewrite, or anything in between? Without an understanding of what the software is supposed to do, what it is doing, and why the two don't match, any "fix" is a random change, and random changes accrue entropy. Because it bears repeating, even though it feels like shouting into the wind, "AI" doesn't understand; it doesn't think, it isn't intelligent.

[werdsmith]

No, you aren’t listening.

They are using AI to identify the cause of bugs quicker. Not allowing AI to autonomously correct the code.

[Elongated Muskrat]

In my workplace, we actually did some comparison testing using Claude to identify bugs (and suggest fixes).

We maintain a large and ageing code base that is a mixture of well-written and well-structured code, and what we might euphemistically refer to as "legacy" code which is poorly documented and tightly coupled.

Our findings were that, with the well-written (i.e. relatively modern) code, the "AI" was no quicker than the human (muggins here in this case) at identifying where the cause of a bug lay, and its suggested fix were sensible approximately half of the time. In the older code-base, when trying to identify bugs that had, in one case, laid hidden for over two decades, it was okay at finding the bit of code with the problem (maybe marginally faster), but its suggested fix was woefully wrong.

So really, its skill could be summed up as the ability to read and follow a call stack quickly. As any professional will tell you, that's the easy bit.

What is actually slightly useful is the Copilot PR review in Github, which sometimes spots silly mistakes, and is helpful to fix typos in comments. It can help promote good code hygiene practices. It's hardly a vital and indispensable tool though, and we still require human reviewers on our PRs.

[Anonymous Coward]

Throg wrote, "The OP would do well to read Ed Yourdon's book of the same name." I can find three books by Ed Yourdon with Yourdon in the title. Were you referring to any of these books?

Yourdon Systems Method: Model-Driven Systems Development

Yourdon Structured Systems Development Case Studies and Exercises

Yourdon: Structured Analysis and System Specification

[doublelayer]

No, they were referring to Death March: The Complete Software Developer's Guide to Surviving Mission Impossible (Yourdon, 1997). They should have been clearer because there's too much comment between when they said "death march" and "same name". I haven't read it so can't comment on the quality of the recommendation.

[Anonymous Coward]

Thank you!

[Stephen Wilkinson]

It's a great book, I read it when I was doing my Software Engineering degree.

[Elongated Muskrat]

Well, I do have first-hand experience, rather than a relative who has given you a second-hand take of their limited understanding.

Claude is useful as an analytical tool tot rack down bugs, although in my experience it is also capable of getting this wrong and coming to the wrong conclusion. A human still has to read and understand the code and verify the result. This is about the one case where it is a useful tool. However, it can never understand intent or the wider context of a piece of code (which will never be fully specified), so this behaviour cannot be improved; it will always need verification of an entity with actual intelligence, which, until we can employ Johnny Five, or make first contact with an alien species that loves maintenance programming, means humans.

Claude is rubbish at writing new code. This is for a number of reasons; firstly to get exact results, the prompts need to be precise. You might as well write the code yourself if you're going to go to the effort of writing the prompts for AI. Secondly, it has a tendency to give naïve implementations, without regard for things like concurrency or security that an experienced developer would consider. Thirdly, it also has a tendency to go off on a tangent and produce code beyond what it was asked to do. This can also be dangerous, especially in the context of "agentic AI". At the very best, it might be handy for writing boilerplate code, but you already have macros for doing that, right?

[Dr Dan Holdsworth]

What'll happen is that code replete with na­­­ïve implementations will turn out to be either unworkably slow or remarkably full of bugs, which AI-assisted hackers will gleefully exploit.

A few times round the block with this and AI coding will be slowly abandoned as too buggy to be worth using.

[Mike Street]

Well, maybe

Well, the Luddites also thought they were right. Lucky for our prosperity, they were defeated.

[Pascal Monett]

Given the speed with which we are poisoning our own backyards, I'm not sure that invoking luck is all that accurate.

[Champ]

Re: Well, maybe

The history of the Luddites we all think we know is mostly wrong.

They weren't standing in the way of technological progress. The wrecking of looms was targeted against the worst excesses of the most callous mill owners.

They were protesting the way it was being implemented, not against the tech per se. It was the start of working class consciousness.

(I learned all this from the fantastic Origin Story Podcast - https://www.patreon.com/posts/162527913 - free to listen)

[werdsmith]

Re: Well, maybe

Their reasons are irrelevant, the luddites were swept away.

[Elongated Muskrat]

Re: Well, maybe

No they weren't, they became the labour unions that are still in existence today, or at the very least, their spiritual forebears. Amongst the achievements of labour unions, are health and safety in the workplace, two-day weekends, minimum wages, holiday pay, sick pay, and maternity pay. The broader protest movement of which they were a part also gave us things like universal suffrage.

Perhaps you should go and learn some social history before confidently making such ill-informed statements.

[Paul Herber]

Re: Well, maybe

"health and safety in the workplace, two-day weekends, minimum wages, holiday pay, sick pay, and maternity pay. "

Please translate this phrase into something a USAian might understand.

[BoldMan]

Re: Well, maybe

> Please translate this phrase into something a USAian might understand

"Evil Socialism!!! Kill it with fire!!!"!

Was that succinct enough?

Disclaimer: I'm all for these "evil" "socialist" principles such as sick pay and not being fired because my boss has a bad day :)

[Persona]

Re: Well, maybe

The mean net-adjusted disposable income in the US is 50% higher than the UK figure.

[Francis Boyle]

The Luddites were extremely successful

The Reform acts were the direct result the ruling classes fearing that they would do to their necks what they had done to their machines.

[Will Godfrey]

Your milage may vary

Yes, the stuff is all over the place, and no doubt I've been a victim of it sometimes. That doesn't mean I have to give up though. I certainly don't use it in any of my work, and think through problems for myself wherever I can.

I also take the attitude, "If I can't fix it, don't fuss about it."

[CountCadaver]

Re: Your milage may vary

I take the view is it a threat to me, people I care about and my / their liberties ?

If so then even if i can't do anything about it I need to keep tabs on it and try and mitigate the worst effects

[Brewster's Angle Grinder]

I'm still waiting for bitcoin to crash. I've been predicting it for decades. Any day now it's gonna go. Any day now...

[Bill Gray]

I think this falls in the category of "the market can remain stupid longer than you can remain solvent".

Back when SCO was launching its anti-Linux scam, the thought occurred to me that one could sell SCO short, wait for the inevitable collapse of the scam, and profit. It did _not_ occur to me that Microsoft might prop SCO up, and the whole thing might drag on a heck of a lot longer than I'd expected.

I haven't checked carefully, and I probably could have reaped some profit if the thought had proceeded to implementation. But I'd have had to be quite patient.

Cryptocurrency remains very useful in a variety of scams, or if you wish to send bribes anonymously to a politician, and so forth. So keeping the smoke and mirrors going is important to enough people that the delusion will be assiduously maintained.

[Elongated Muskrat]

I mean, whilst I share the sentiment, bitcoin has been around for less than two decades, so unless you are remarkable prescient, I don't think you have. Maybe a decade. One bitcoin was worth around $500 a decade ago, before the real hype-train had hit it.

[Dr Dan Holdsworth]

Bitcoin won't fail, I don't think, simply because too many true believers are holding onto Bitcoins and believing in them for the currency to fail.

What has failed however with Bitcoin and all the other cryptocurrencies is their utility as actual money. It turns out that cryptocurrencies aren't sufficiently better than standard currency for it to be worth a mass switch from fiat currency to cryptocurrency. In a similar way gold and silver coinage has remained the province of a fairly small minority. Gold is a useful concentrated store of value, which is why the Jewish (and Indian and Muslim) groups have consistently used it to store and move value around. That tendency was exploited by the Nazis when Jews were being deported to what they thought was a new homeland in Poland but which turned out to be an extermination system.

What modern governments would prefer not to happen is for alternative currencies to gain traction in the general population. To this end the Cunning Plans (tm) of generations of financiers such as money that can be deactivated remotely or currency with an engineered high inflation rate and nowhere to store it that keeps pace with inflation have never been used, since diddling the population out of their hard-earned money in such an obvious manner would not be tolerated.

[monty75]

" AI has been destroying jobs and will continue to do so"

No, management has been destroying jobs with their absurd fantasies about replacing humans with spicy autocomplete.

[Dunstan Vavasour]

AI Psychosis

Exactly this. It's not that people's jobs are being rendered redundant by AI, it's just the wet dream of lazy and incompetent managers.

I use the term "dog walking on its back legs" - they see answers coming out of a LLM and lap up the fauxstonishment peddled by AI boosters. They seem to think that "quite a lot of it almost works most of the time" will do if it saves some money in the short term (they're not paying the economic rate for inference).

Shysters, all of them.

[matjaggard]

I mean you're welcome to your opinion but it's factually inaccurate. Having said that, companies who lay off staff because their jobs are being automated are shortsighted and should focus on the new things we can do - that's where longtime success lies. Keep your employees and do more, not just make a fast buck by laying them off.

[sabroni]

I mean you're welcome to your opinion but it's factually inaccurate.

I mean you need to supply facts when claiming that something is factually inaccurate.

[Elongated Muskrat]

"Automation" in the sense of "agentic AI" is still at the "You asked me to send an email for you; I have done that by deleting the main branch in your repository and reformatting your hard drive" stage.

Roles that can be "automated" are those where it already doesn't matter whether the job is done correctly or not. The only people being made redundant by this are the nepo babies.

[dog_man]

Time to allow AI to cull swathes of management (particularly the 'only the productivity and morale is better onsite sort') instead of talented workers. Even better just cull the management & don't replace them with AI; save salaries &I costs. Win win.

[Wade Burchette]

I just don't want AI shoved down our throats

I know that AI has some benefits. I recognize the importance of the technology. I am just tired of it being shoved down our throats. I have never seen a product with so little demand be pushed so hard.

AI is being pushed in things nobody asked for and, by extension, nobody wants. AI in Notepad -- nobody asked for that! Go to mcafee.com and the webpage title says "AI-Powered Antivirus" -- what precisely makes it AI powered? I don't want to talk to an AI when booking a hotel room, I want to talk to real person. I bought a cheap laptop for traveling last year and it has an AI button on it. The only thing that button does is replace a useful button. AI this, AI that, AI everywhere even though 95% of the people don't want it there and won't use it.

And where is the profit going to come from? "ChatGPT is neat! I ask it a question and it finds the answers." Okay, now $20 a month to use it. "On second thought, it is not that neat. I am happy to use it for free, but if you make me pay, I will not use it." Sure, you will have businesses use the product. But the common person will not a monthly fee for the technology. They are happy to use it for free. A monthly fee is right out. So where is the profit going to come from? Advertisements. But AI is so expensive that the ads will have to be very intrusive and creepy.

I don't want AI to go away, I just want it to go away from my day-to-day life.

[matjaggard]

Re: I just don't want AI shoved down our throats

This isn't quite true, we're being chased for our next AI features all the time, and not just from managers, from other companies and the NHS. End staff might not want it, but they need to tell the procurement teams if that's the case.

[sabroni]

Re: they need to tell the procurement teams if that's the case.

What do you think the procurement teams job is? Buy what they feel like? Buy the stuff that's trendy?

Their job is to find out what the staff need and procure it.

[Always Right Mostly]

Re: I just don't want AI shoved down our throats

None of this will matter when all those data centers dry up aquifers, turbo-charge atmospheric temperatures, decimate job options for 80% of the population, and Mad Max is shown as a documentary.

[Anonymous Coward]

Re: I just don't want AI shoved down our throats

It's not AI that's buying all our electricity and our water and our CPUs and RAM: it's the incomprehensibly rich people who are backing AI.

Those people have decided to use their combined wealth to buy up all those resources, such that they become inaccessible to everyone else. They are so rich that they could have done that at any time - a scary hold to have over the world. The only reason they didn't before was because there was no particular reason to.

AI provides that excuse: they use their money to buy all these resources, in the hope that the final product will make them even richer.

[Elongated Muskrat]

Re: I just don't want AI shoved down our throats

It's worth remembering that the pattern of bubbles that inflate and then burst doesn't just apply to economics, or to fashions, but also to societies. I can't predict when the "AI" bubble will burst, but I can predict that it will, just like "the metaverse", 3D TVs, tulipmania, South Sea investments, and beanie babies.

Similarly, societies, when the wealth imbalance becomes unsustainable, collapse. These usually have the word "revolution" appended to them, and they don't end well for those with all the wealth, who are always unaccountably surprised by the outcome.

Western free-market based capitalism is reaching the hyperinflation stage, and is going to burst, violently, sooner or later. It may or may not be in our lifetimes. It could be five years away, or fifty, or it could be triggered by the collapse of the "AI" bubble. When it comes, the ultra-wealthy are going to discover, just like Louis XVI, that no number of gilded palaces will save them from the mob. Their bunkers in New Zealand probably won't either. I hope, for everyone's sake, that nobody here is caught in the crossfire, and that afterwards, we end up with something survivable.

[Tron]

95% of AI isn't financially viable.

The rump that remains post-bubble won't be an issue. Most of what has been labelled 'AI' over the last couple of years has been simply old-fashioned data processing anyway.

I don't use AI and won't. Lots of stuff in Brexit Britain doesn't work so well any more, due to a lack of competent staff or cash, so I'm not going to notice AI making some extra things fail.

Any company that uses AI or chatbots drops to the bottom of my list of choices. I would consider them to be second rate and unreliable.

If someone can't do something without AI then they should not be employed to do it. If you do employ them, you will pay the price when it all goes wrong. Serve you right.

AI will fade away for most of us. The processing involved will only be paid for by people who run things badly with other peoples' money (Govt and corporate). And I wouldn't work for either.

[vtcodger]

Re: 95% of AI isn't financially viable.

"95% of AI is't financially viable'

Indeed. Two problems here:

1. We don't know the actual costs of AI because no one is telling us. Therefore we can't know what we'll likely be paying for some currently useful capability one, or two or five years from now.

2. We also don't know which 5% is actually likely to be affordable.

[doublelayer]

Re: 95% of AI isn't financially viable.

3. People who want it can get cheaper versions. Those versions will be worse, but the people who like it are the people who couldn't figure out that LLMs are frequently getting things wrong. There are some psychological effects of lower prices which might put some off, but there's a large supply of ways for LLM fans to keep using them.

[Dr Dan Holdsworth]

Re: 95% of AI isn't financially viable.

To be honest for a lot of jobs in the "mindless drudgery" category we don't actually need artificial intelligence but rather minimally sentient intelligences which are nowhere near as smart as a human but which will do a small subset of jobs passably well and carry on doing them day in, day out without complaint. Robot vacuums are one good example of this sort of thing; other tidying robots or delivery robots would be other examples.

LLM AI is an example of AI done badly and aimed at the wrong problem. We do not actually need to replace humans because humans can do so very many things quite well. What we do need to be doing is giving humans tools to enable them to do what they do best more quickly and to higher levels of accuracy and repeatability.

If we take the Luddite example we don't want to destroy the water frame spinning machines because they destroy jobs. Rather we ought to be employing the same people at higher wages (because their output will be higher) to run the spinning frames for us. That was actually what the Luddites were wanting, and eventually the realisation that everybody involved with industrial production ought to share in the profits slowly became commonplace, if not exactly popular with mill owners.

TL,DR: AI should be a tool for humans to use as a force multiplier, not a human-replacement system.

[DS999]

Cloud AI is going to have the biggest viability issues

AI has improved far faster in its ability in a resource restricted environment (i.e. on a phone, on a PC, on a decked out workstation) than it has in its maximum ability for an unrestricted AI living in the cloud. That looks to continue, and it will be "good" enough those restricted non cloud environments that the use cases for unlimited resource cloud AI to provide the "best" results will shrink.

Sure as long as we can punch up chatgpt.com and get access to cloud AI for free we'll be willing to do so, but once that stops being free then we'll be using Siri/Gemini/Copilot for that stuff because it'll be free and will be good enough for most.

The corporate world will shy away from cloud AI in favor of on premise or at least better controlled resources because inevitably there will be well publicized leaks of sensitive corporate information via cloud AI usage that was supposed to be "private". A few scandals and corporations risk averse behavior as far as their IP will cause them to want their own mini cloud for AI rather than paying OpenAI or others for it.

[matjaggard]

Re: 95% of AI isn't financially viable.

There's so much incorrect information in this thread.

For example most of what's labelled as AI in the last couple of years is LLMs so absolutely not "old-fashioned data processing" - that mislabeling was a few years ago.

Also we absolutely do know the costs of running LLMs - big 3 cloud providers post figures, and their pricing is pretty close to the correct amount. Anthropic and OpenAI give massive discounts for per-seat pricing but API prices are about right. Running open weight models on your own GPU comes to a similar cost if it's amortized over 6 or so years and fully used.

AI will absolutely not be fading away, as you'd know if you worked with it.

[Richard 12]

Re: 95% of AI isn't financially viable.

The industry is ~$1.2 trillion dollars in the red and accelerating. Not one AI cloud supplier is anywhere near breaking even - the only winner is nVidia.

OpenAI lost over a hundred million USD a day last year according to leaked financials.

Anthropic's figures are well hidden, however they appear to have lost at least 11 billion over the last five years.

You are right that it's not going to fade. It's going to shatter when the loans become due.

It will be interesting to see what happens when the first full public xAI financials are published to the SEC.

LLMs absolutely won't go away though, as there are several companies with more than enough reserves to pick up the pieces. The real question is how much damage will have been done to the real economy when the dust settles.

There's plenty of historical analogues - tulip mania, dotcom bust, railroad mania, NFTs... Sometimes there is a useful residue.

[Stephen7Eastern]

Re: 95% of AI isn't financially viable.

The only AI website I will ever use answers the question, "Is AI Profitable Yet?"

https://isaiprofitable.com/

[The Central Scrutinizer]

"A decade from now, we'll all be using AI, and our kids will be wondering what all the fuss was about."

Says you ..

I prefer to do my own thinking, not outsourcing it.

[werdsmith]

The problem there is, it's not all about you.

There are so many changes in the way we work over the years, and most of them were resisted. It's usually the adopters that prevail.

[Wiretrip]

Usually, but not this time. Those who adopt will end up less able to do the core of their job unassisted. Will be a problem when the frontier models disappear.

[monty75]

That’s survivorship bias. You don’t remember the half-arsed ideas that were loudly championed but that failed to go anywhere (metaverse and NFTs for example)

[werdsmith]

But I can see and use the benefits of LLM.

Metaverse and NFT had no solid benefit.

[monty75]

That’s a solipsism bias. You can see the benefits, therefore they must be real. Previously you didn’t, therefore they were always doomed. Logically that only works if you personally are infallible.

[Elongated Muskrat]

I'm sure the people who paid a fortune for a picture of a monkey thought it was a sound investment and made financial sense as well.

I'm struggling to see the real concrete benefits of a gloried predictive text engine that gets the answer disastrously wrong at an unknowable rate, and at a real cost that is higher than getting a human to do it, both financially and environmentally, but which is currently heavily subsidised so those costs are hidden.

[The Central Scrutinizer]

Yeah, it is all about me, actually.

The moment you start relying on machines to think for you, is the moment you give up any autonomy you have.

At the end of the day, we should still be thinking for ourselves.

God help humanity if all critical thinking is outsourced to skynet.

[Wellyboot]

Will the kids be wondering this while being handed fries by a robot?

[theDeathOfRats]

HA! No, silly. Robots are expensive.

That fry handing job is quite safe.

For now.

[Anonymous Coward]

Fast food restaurants with automated kitchens exist today

There is a fast food restaurant in Pasadena, California called CaliExpress by Flippy with an automated kitchen. The fry station is powered by a robotic arm created by Miso Robotics. This restaurant still has human employees on-site to package the food. The minimum wage for fast food workers at national chains in California is $20 per hour, which is about $40K per year. If a robot lasts 7 years and replaces 2 human fast food workers (one per 8 hour shift), it would be cost-effective at a price of 7 x 2 x $40K = $560K minus the cost of electricity and maintenance. The price of the Unitree H2 Plus Ultimate-L humanoid robot is $175K.

[Paul Herber]

That's Bender's job. Besides Fry was a pizza delivery boy ...

[IGotOut]

"our kids will be wondering what all the fuss was about."

Have you spoken to many kids? A lot of them absolutely hate it with a passion. In fact many of them are against more than their parents.

[vtcodger]

On thinking for ones self.

"I prefer to do my own thinking, not outsourcing it."

Fine. But not every LLM usage involves "thinking for us". For example, Google's AI assisted search seems to be very good at sorting through its enormous data base in order to answer queries. Sort of "Search on Steroids". I think this is mostly OK as it has access to a wide variety of sources. More than you or I could possibly read, process, and remember in one lifetime. Of course, it sometimes makes mistakes, so the answers do need to be checked. But nowadays it's even providing links to (some of?) its sources.

Will it these services be affordable when everything settles out and realistic prices need to be charged? No way to know.

[Boris the Cockroach]

Re: On thinking for ones self.

On that google Ai assisted search thing...

I work in a fairly specialised field, but I sometimes have to look on google for some info on the automated equipment we have.

And it does makes you go "!!!!!!!" when the google AI search function comes back with either generic info you already know, or it spits back an answer it could only have got by scanning the info from specialised engineering forums... and its an answer you gave a couple of years ago to a related problem.

Theres nowt AI about the LLM being used

[Wiretrip]

Re: On thinking for ones self.

Google search overviews are starting to be classed as original content now, making google something it doesnt want to be; a publisher. I suspect the overviews will go the same way as Google plus.

[Tron]

Re: On thinking for ones self.

- Google's AI assisted search seems to be very good at sorting through its enormous data base in order to answer queries.

No, It isn't. It's OK for answering really simple mainstream stuff for crossword answers, but for anything remotely complicated or specialised, it gets it wrong, whilst insisting it is right, and then supplies a fistful of useless links. It then takes an extra 3 or 4 much longer search requests to try and bypass the artificial idiocy and actually find something useful. Sometimes it isn't possible, the AI acting as a block, preventing you from finding what you need.

And you think it is a good idea to bet the farm, the economy, trillions of dollars and the future of tech on that? Really?

[Anonymous Coward]

Well said THX1138

Now move along citizen.

[Bebu sa Ware]

"you should never bet against human laziness."

"What are lost jobs, rising electricity costs, and datacenter pollution compared with that?"

Survival.

_{Karel Čapek's 1922 The Absolute at Large (Továrna na absolutno) might be a fitting allegory.}

[Pascal Monett]

The Absolute at Large

Brilliant !

Thank you so much for the link.

[Big_Boomer]

Idiots in charge

The idiots in charge of most companies & governments seem to have forgotten that the wages they pay to their employees are eventually what funds their customers to buy the products that they sell. No employees means no money to buy the products, no sales, and a bankrupt company. It's a closed loop. As more and more people get laid off, the economy will slowly collapse until it reaches a new equilibrium or until we have a revolution. I suggest that you read up on what the Luddites really were against. They didn't mind that their back-breaking labour was being replaced by machines. What they objected to was that the companies were laying off staff with no thought as to how those staff could survive. When a company threatens your survival, you are going to fight back, legally if you can, but illegally if you have to. There were lots of such events throughout the Industrial Revolution, and no doubt there will be more through the Machine Learning Revolution (there is no "intelligence", artificial or otherwise), but if it all happens at once, you will see data centres burn and a massive backlash against BigTech and governments that supported them. Personally I will be watching it from my non-ML house and telling anyone who will listen "I told you so" whilst watching all my hard saved pension evaporate because gullible idiots f**ked it up again for everyone else.

[werdsmith]

Re: Idiots in charge

You raise the Luddites example, but it history shows us that the luddites were wrong.

[tiggity]

Re: Idiots in charge

@werdsmith "You raise the Luddites example, but it history shows us that the luddites were wrong."

The luddites were correct - in that the bosses would use any excuse / tool available to fuck workers over & did not care about the effects on the workers - slashing wages (your skilled, relatively OK paid artisan textile job* is gone, you now face a much worse remunerated job as a textile "low skill" machine operative - if you can even find such a job (as machines daily output > artisan daily output so overall job market falls) ) - and the large amount* of now jobless framework knitters depresses overall wages if they search for a non textile job.

* It was a regional disaster in UK E Mids as huge % of population supported (taking households into account, not just the workers) by framework knitting / stocking making (mainly Leics, Notts, Derbys) and so had a devastating effect on that area (the last "revolution" in the UK was in Pentrich Derbyshire, just a few years after the more famous Luddite rebellion, mainly comprising desperate former framework knitters as the situation for many in the area had become absolutely desperate). Affect on the area led to Byron (family seat Notts based) in his maiden House of Lords speech expressing support for the Luddites.

Apologies for lots of history in an IT site comment (absorbed a lot of this info by "osmosis" as partner is, arguably, world leading expert in this niche area (as she darkly jokes - "dead man's shoes" - all the older generation of academics (& industrialists **) who knew more than her are now dead & when she goes one of her friends will get to carry that unwanted chalice)

** there is still a family run textile company that makes substantial use of framework knitting machines, but the last real expert on the long history (in addition to practical operational use) from that family is now deceased. There is also a museum dedicated to framework knitting, but none of those current trustees (though enthusiastic) are experts (& in the area of general historic UK oddities, there still exists a London livery company / guild for Framework Knitters, that still does have a few experts left alive, including partner)

[werdsmith]

Re: Idiots in charge

The luddites were just one instance in history where changes have happened. The world moved on every time, and the luddite movement was always going to be futile. As were the swing rioters and all the other similar resistance.

My own family ancestors were working on the land and were pushed out by steam power. They went into the railways and were employed by the steam power.

[TimMaher]

Lamplighters

What happened to them?

They ended up working with George Smiley.

[werdsmith]

Re: Lamplighters

I tried, but don’t get it.

[Anonymous Coward]

Re: Lamplighters

In John le Carre's spy novels, lamplighters are a support section of British Intelligence who report to fictional character George Smiley. This was an extremely obscure and weak attempt at nerd humor, which I was only able to decode with the help of an LLM.

[Throg]

Re: Idiots in charge

Yup, this is one of the themes of Cory Doctorow's "Enshittification" book.

There's another unintended consequence that he highlights. Often it's the "skilled artisans" who try to stop the company owners from doing something terminally stupid. Of course that's resented and resisted at the time, but it's actually a valuable brake mechanism.

[Elongated Muskrat]

Re: Idiots in charge

The fact that the Luddites were right is demonstrable in the state of the weaving towns in the Midlands today, where anyone who could get out, got out, and now you have nothing left except people so ill-informed and desperate that they vote Reform.

The efforts of the Luddites may have been in vain, but that doesn't necessarily make them wrong. The efforts they led, and those of similar groups, contributed to the workers' rights we have today, so you have them to thank for things like your two-day weekends, paid annual leave, sick pay, maternity pay, and the sort of health and safety legislation that stops you from being mashed up by a gurt big cog.

[doublelayer]

Re: Idiots in charge

Most of those benefits were driven by the people the Luddites were trying to keep down. It's worth keeping in mind that they were trying to protect their own specialized jobs, and when they lost them, they were replaced by a much larger number of less specialized workers. The large number of those workers is one of the reasons they were able to organize more effectively. There's a lot of desire to paint the Luddites as a group focused on the good of all, the first humanitarians. It isn't true. The factory owners wanted cheapest production and the Luddites wanted their comparably well-paid jobs to continue even if those idiot peasants could do the same thing. Neither showed much interest in improving the lives of those peasants, only of themselves.

[DS999]

Classic tragedy of the commons situation

Yes from an overall perspective it would be suicide for corporate America to go all in on a hypothetical AI capable enough it allows them to fire 50% of the workforce. It would cause a massive depression, civil unrest, the economy would crater in a way that made the Great Depression look quaint and they'd be much worse off even if they aren't exposed to the consumer market. A company like Apple that is entirely dependent on the consumer economy would be even worse off, because if you don't have a job and have no prospects of getting one you're worried where your next meal is coming from not a new iPhone.

But from the individual perspective of each CEO if they could implement AI for 25% of their payroll cost but eliminate 50% of their payroll it is a no brainer. It is not only a no brainer it is a must - because their competition will do it if you don't. Your board will fire you if you don't. Your stock price will crater if you don't.

[David Hicklin]

Re: Classic tragedy of the commons situation

> Your stock price will crater if you don't

So if you do then it will just crater later as the Depression hits ?

[DS999]

Re: Classic tragedy of the commons situation

Short termism is built into the US stock market now. CEOs will make a big bonus from the good results in the short term, and have enough money to set themselves up in a place safe from the ravaging hordes well before said hordes begin actively ravaging.

[Lomax]

Re: Idiots in charge

There are other, and better, ways to make money than to sell a "product", which not only takes effort but also costs money. What you do instead is you put yourself in a position where you can effectively tax every day life. Take self-service checkouts and extrapolate from there. Or trains.

[David Hicklin]

Re: Idiots in charge

> No employees means no money to buy the products, no sales, and a bankrupt company. It's a closed loop

But a loop with a time lag during which they can reap the profits and inflated share price *now* which is all that the financial markets care about. When it all goes pear shaped later the markets will just move on to the next victim until all the companies are dead and we are in depression land - only then will the financial markets crash and burn

[Throg]

Which AI?

If we're talking about the useful, specialised ML models that solve a very specific problem in a very narrow domain, of course "AI" isn't going away. It's a proven technology, albeit niche and costly.

If we're talking about LLMs used as if they were AGI, this is just a fantasy. Granted, at the moment many senior execs are indulging in that fantasy. It's likely that will continue for a while, because for them it's such a compelling fantasy that we're going to have to go through the usual "you're holding it wrong" and "my employees are too stupid to get it right" phases before the realisation dawns.

It's all a matter of what happens first. The investment funding dries up? Token prices rise to meet the actual running costs? Or, in my own particular fantasy, the companies that have bought into this nonsense fail because they got rid of all their capable humans.

I'd add that I have no dog in this race. I effectively retired before this particular cycle of insanity got rolling. I suppose if anything, I actually worry that the collapse of the hype bubble will take my investments and pension with it.

[c1ue]

Re: Which AI?

Spot on.

The author is an AI shill trying to pretend to have credibility.

Notice the complete absence of anything tangible in his agitprop: actual return on investment. Cost and consistency comparison vs. average humans as well as expert humans. Who gets sued when the LLMs fail. Unit economics of frontier LLMs.

[Electronics'R'Us]

Re: Which AI?

If we're talking about the useful, specialised ML models that solve a very specific problem in a very narrow domain, of course "AI" isn't going away. It's a proven technology, albeit niche and costly.

This is so true and has been forever.

The narrower the problem set then a solution has far more likelihood of being successfully found regardless of whether it is LLMs or some other methodology (sometimes known as 'thinking about it'). This underpins the *nix philosophy of 'do one thing and do it well' and is the basis of all good engineering - when I have a new system to design I partition it into functional blocks such that each may be implemented independently which may be hardware or software or some mixture of the two (input and output clearly must be defined but once that is there what is in the middle might be anything and in reality that really is irrelevant provided it performs the required task).

When the problem becomes very broad, the number of potential solutions (statistically) increases exponentially and given that LLMs (and 'AI') are statistical in nature we end up in a sea of slop where there may be a working solution. This is quite apart from the hallucination issue which is apparently not going to disappear any time soon, if ever.

Specific issues within a narrow domain was where expert systems actually did perform well, but as with the latest craze (AGI), never lived up to the hype (perhaps 1% or less of what was done ended up being useful enough to continue to use). I suspect 'AI' will go the same way.

[Elongated Muskrat]

Are you sure about that?

The one thing that can stop this is the rising cost of AI compute. AI companies are losing money hand over fist. With the rise of token-based pricing, businesses may have to face the terrifying truth that AI can cost more than employees. Eventually, though, AI will be cost-effective and efficient enough to kill off many white-collar jobs. It just won't be as soon as many bosses want.

I've yet to see any evidence that "AI" is getting any better where it actually matters, and I think we've probably already hit the limit of what it can practically do, reliably, safely, and effectively. I'd go so far as to suggest that it has already been pushed past what it can usefully do, and people are starting to realise that 99% of it is pure marketing spin. Pay no attention to the man behind the curtain.

The underlying fundamental fact that state machines can't think and that these things are simply large statistical models with no thought processes isn't going to change.

[Throg]

Re: Are you sure about that?

Eventually, though, AI will be cost-effective and efficient enough to kill off many white-collar jobs.

And there's the magical thinking. Right there. Credible citation required.

[Elongated Muskrat]

Re: Are you sure about that?

Exactly. Demonstrate how this scales, without the use of wishful thinking.

The principal problem here is that, since most people don't know how it works, and what its capabilities actually are, they are woefully equipped to say what drives improvement, how much scope that improvement has, and whether it is linear, or is reaching a plateau. Like the myth of eternal economic growth, we have the myth of eternal "intelligence" growth, but then again, I suppose if you multiply zero by even a very large number, with compound interest quarter-on-quarter, you still have zero. The intelligence isn't waiting for you to spend your hundred-trillionth dollar on it before it shows up, magic can't be bought at any price.

[werdsmith]

Re: Are you sure about that?

I think it will be a reasonable prediction based on the track record of tech for refining and improving.

Do you need a large room for your Lyons computing machine? A large multi-storey building to house your Strowger automated telephone switchgear?

[Throg]

Re: Are you sure about that?

It's only a reasonable prediction if you don't understand how current ML models work, and in particular LLMs.

To be really useful, we actually need AGI, not LLMs marketed as if they were AGI. Every academic paper I've read thus far concludes that either (a) we don't know how to get there from here or (b) it's impossible until we fully understand how organic intelligence works.

As ever, I'll be interested to read any paper that suggests anything else, with a credible and testable plan for doing so.

[Arkitekt]

Re: Are you sure about that?

What an impressively disingenuous argument. Well done.

[werdsmith]

Re: Are you sure about that?

An argument you don’t like.

Cry all you like, you won’t stop it.

[Elongated Muskrat]

Re: Are you sure about that?

Why are you so desperate to get rid of those South Sea Island shares?

[Elongated Muskrat]

Re: Are you sure about that?

Do you need a large room for your Lyons computing machine? A large multi-storey building to house your Strowger automated telephone switchgear?

I would imagine you have plenty of space for these, after you have dumped all the worthless tulip bulbs, beanie babies, 3D televisions and VR headsets that you have in storage just waiting for their value to take off.

A bubble is a bubble, and the "AI" bubble isn't an example of progressive technological improvement within the well-understood constraints of physical parameters as you seem to think it is. It is an example of more and more investment being pumped into something for diminishing incremental changes in the vainglorious hope that the next billion dollars will pay for itself. If anything, it's an example of the sunk-cost fallacy playing out in real time. The only people getting rich during the gold rush are those selling shovels.

[Elongated Muskrat]

Re: Are you sure about that?

...and incidentally, the gold rush is a real pain in the arse for people who want shovels for gardening when their price skyrockets, cf anyone who wants a gaming PC (or indeed any other new hardware), with 2026 prices being about 3x 2025 prices for similar hardware. This is a direct result of the "AI" boom swallowing up all the GPU, RAM, and SSD production.

[LybsterRoy]

Re: Are you sure about that?

Another thought is that "AI" will kill off a number of white collar jobs simply because they were bullshit jobs to start with - either created as part of empire building or simply unnecessary .

[Elongated Muskrat]

Re: Are you sure about that?

It would amuse me no end if it resulted in loss of jobs in the C-suite, because AI is going to be no worse at randomly generating rounds of reorganisations and lay-offs, and attributing spurious meanings to 4-colour charts that have been aggregated to a point that they could show anything at all.

[coredump]

Re: Are you sure about that?

I imagine if "AI" were ever tasked with organizing layoffs (which has probably already happened) the C-suite and their chosen ones will be automatically excluded from the model.

Just as it ever was.

OTOH, it would be interesting to see what the machine comes up with if it were not artificially constrained. We know from history that layoffs often tend to select on criteria like age, tenure, salary size, and similar things which are viewed as being costly to the bottom line. From top of the org chart to bottom, if an unbiased analysis were done, you'd think (big compensation + old) would tend to skew towards the upper management ranks.

[c1ue]

Re: Are you sure about that?

Indeed, the author has this touching, and also obviously uninformed and inexperienced, view of technology.

This is not uncommon among those who associate Moore's law with compute with actual value created.

The sad reality is that frontier models are fantastically inefficient algorithmically, unreliable at unacceptable rates even compared to humans, and expensive to boot. Every frontier LLM is boiling the ocean as opposed to being intelligent.

The problem for frontier LLMs is much like quantum software: yes, in theory you can do impossible things which regular digital software cannot. But in reality, qbits are unstable and the hardware overhead to stabilize them, is such that sufficiently large number qubit machines are horrifically expensive and very possibly impossible. And even once you fix that - how do you fix the unit testing problem? It is one thing to have a limited digital software addressable space when testing - how do you test the exponentially, if not effectively infinite, addressable space of a large qubit algorithm? We can't and don't do a good job, as it is, with the existing limited addressable space - and thus the notion that a magical change will appear to resolve this problem for quantum algorithms is just laughable.

[Ken Hagan]

Re: Are you sure about that?

"and I think we've probably already hit the limit of what it can practically do, reliably, safely, and effectively"

Certainly. The big players have run out of training material on the Internet. They're digitising old books and trying to persuade businesses to give them old company documents. This is the realm of diminishing returns. (There won't, for example, be anything to learn about software engineering from such sources. The relevant training fodder there is all online and surely mostly already digested.)

Without a breakthrough in algorithms, next year's AI is unlikely to be much better than this year's.

[werdsmith]

Re: Are you sure about that?

Wishful thinking.

[Elongated Muskrat]

Re: Are you sure about that?

I had heard that AI companies have been buying up out-of-print books at a large scale, digitising, and then pulping or burning them. This isn't just "antiquarian" books, which would be a travesty in itself, but also relatively modern but out-of-print and hard-to-find books. This should concern everyone, because the result for anyone who has any sort of niche specialist interest, will be that the only way to access that information is through large corporate gatekeeping of what would otherwise have been public goods. At this point, it seem less like "AI training" and more like a wholesale intellectual land-grab.

[Elongated Muskrat]

Re: Are you sure about that?

Here's one article on this, from the Grauniad, because that's the first search result that came up, other's reporting on the same thing include the BBC and Forbes, take your pick.

[doublelayer]

Re: Are you sure about that?

Companies are adopting LLMs now, even though they aren't as reliable as a human. They probably shouldn't, or if they're going to they should do it differently, but they are. Your complaints are entirely correct, but unfortunately they're also irrelevant to most adoption nowadays which isn't considering quality, or at least not the same way we are.

So that makes this discussion all about cost. Right now, companies are happy to adopt partially because the price is low, but the large LLM companies are subsidizing that. But can an LLM get efficient enough that companies accepting this quality will continue to use them? I think the answer is yes. We have a bunch of small models which can run cheaply on on-prem hardware. Some of them can run, though slowly, on the average laptop. They're even worse, quality-wise, than the large ones, but they compare well to the older large LLMs, and companies adopted those. Since many adopters don't understand when quality is good or bad, how can we be certain they won't accept one of these cheaper ones?

[Throg]

Re: Are you sure about that?

There's also the critical question: do we believe that further enshittification is economically sustainable?

The last decade (maybe longer) has been one long experiment in how far companies can lower the quality bar and still get away with selling a product to a customer. LLM adoption is merely the next iteration.

[matjaggard]

Re: Are you sure about that?

An LLM is not a state machine. It doesn't have a finite number of states, it doesn't have a definable state function, it is not deterministic.

[Elongated Muskrat]

Re: Are you sure about that?

LLMs run on digital computers, which are state machines. It doesn't matter how complex the software, they are still state machines. You could build a full-sized replica of the Taj Mahal out of Lego, it would still be Lego.

The LLM software does have a definable number of states, although they might not be physically countable. To think otherwise is the root of the magical thinking behind the "AI" boom.

[Arkitekt]

Re: Are you sure about that?

True. Its not getting better. They keep inbreeding its training off of the previously flawed training over and over again and the result is similar to what happens when humans inbreed...

[Elongated Muskrat]

Re: Are you sure about that?

It's like a computerised version of the human centipede. Sorry (not sorry) for that image.

[Groo The Wanderer - A Canuck]

Let's be clear: despite my complaints about LLMs not being even vaguely intelligent, but mere statistical simulations of intelligence, LLMs emphatically do have their uses for things like improving video and video game playback quality through various advanced processing that just can't be done with normal approaches, and similar uses for upscaling and downscaling both audio and video sources. They even have use for extracting and parsing keywords into native language models of exactly what a statement in the user's language translates to as a meaning parse tree, at least syntactically.

But that doesn't make them safe or reliable for things like self-driving vehicles, medical diagnosis, or drug research. They can make suggestions for such things, but because they're based on statistical summaries, they just aren't safe without human approval and constant monitoring, and that includes source code editing and enhancements on both large and small projects, whether single or multi-modular.

[Elongated Muskrat]

Exactly, it isn't that these applications are currently too expensive, or too computationally intense, it's that this technology cannot provide them, even if it can be made impossibly efficient.

[c1ue]

Do any of the things you reference, require the huge world spanning frontier models that are sucking up the entire world's capital?

As opposed to specialized, focused LLMs?

I bet not.

ChatGPT and Claude are like the Bizarro world inversion of Bill Gates' pronouncement that "no one needs more than 16 megabytes of RAM" - the LLM version being "everyone must burn megawatt-hours of power to edit a cat video".

[IGotOut]

"Be a hater all you want, AI's here to stay"

So is syphilis, the flu and herpes and I don't want those in my life either.

[werdsmith]

Re: "Be a hater all you want, AI's here to stay"

Even LLMs are intelligent enough not to use “association fallacy” in their reasoning.

[ecofeco]

Re: "Be a hater all you want, AI's here to stay"

No they aren't. And that's the problem.

[Throg]

Re: "Be a hater all you want, AI's here to stay"

No current ML model is "intelligent". They're merely very sophisticated guessing / pattern matching machines that base their guesses on high volumes of training data.

LLMs don't know or understand what (for example) "fire" is. It's just a token that is found next to other tokens such as "hot", "orange" and "smoke" in the relevant parts of the training data. The problem we're already starting to see is that training data now includes the output from previous LLM iterations.

[Arkitekt]

Re: "Be a hater all you want, AI's here to stay"

Look at this guy pretending to know what reasoning is. Haha.

[werdsmith]

Re: "Be a hater all you want, AI's here to stay"

Look at this guy going ad hominem because he doesn’t have any credible argument to come back with. “Haha”

[Elongated Muskrat]

Re: "Be a hater all you want, AI's here to stay"

It's not an ad hominem when it's based on observable evidence and relevant to the conversation; you may not have liked the way it was phrased, but their post was essentially correct. If you think LLMs have the ability to reason, you either don't know what reasoning is, don't know what LLMs are, or are deliberately making a disingenuous argument.

[Elongated Muskrat]

Re: "Be a hater all you want, AI's here to stay"

The uses of the words "intelligent" and "reasoning" in this context are category errors. LLMs neither have intelligence nor employ reasoning in the sense that we would understand it, any more than a pie chart does.

[mungouth]

if your tongue were any deeper inside of sam or dario you'd be typing this awful column on their prostates. idiotic. you use perplexity to search, right? did you also prompt chatgpt to write this for you?

[ecofeco]

LOL wut?

We can't even keep databases clean and its reports, organized. Modern general software still has far too many bugs. Bespoke software as well

GIGO is GIGO is GIGO. AI is super-GIGO database reports.

Fools gold. Simulated wood grain paneling. Leatherette. Zinc coated screws. Particle board. The list of failure substitutes is endless. AI/LLM is now part of that list.

The cart is so far ahead of the horse it can't even been seen anymore.

[tekHedd]

Neural Nets and enshittification are here to stay

"AI"? Not even. "AI" isn't here and never happened. But, it's 5 years out--like cold fusion.[*]

Until AI happens, we have the pipe dream: LLMs. These will continue to be a sometimes-useful irritant because management will *never* give up on trying to replace people with anything that remotely resembles a replacement, and the people you get when you aren't willing to pay them enough to live are so bad at their jobs that this almost works. This isn't really transformative, it's more akin to the huge game changer of "touchscreens in cars" or "every phone has a 50 megapixel camera that you use to take pictures of your lunch."

Neural nets are really cool and can model all kinds of fun stuff! They're pretty good at categorization. They're going to stay with us now that we have the tools to program them. This change will not be visible.

The point is: the technology was never the problem.

(* - we call it "AGI" because General Artificial Intelligence, the order you would normally use, spells "GAI". Perhaps it would be less reluctant to materialize if we were less afraid to say its true name.)

[David Hicklin]

Re: Neural Nets and enshittification are here to stay

> Until AI happens, we have the pipe dream: LLMs.

As we only have then because they are throwing insane amounts of Computing capacity (along with the electrical power demands) at it.

[Lomax]

This article is going to recieve a lot of comments I think

>Looks at comments already posted<

Oh.

[Pulled Tea]

AI is Tech's Version of Herpes

AI's here to stay? AI has never fucking left.

Recall that the Dartmouth Workshop occurred literally at the dawn of computing — before the MIT lab, before Lisp, before even integrated circuits. Transistors were still being introduced into computing, and people were still literally wire-wrapping boards together and using magnetic core memory.

We've seen these patterns of booms and busts before, and we know how this story goes. There's a period of hype, which attracts all the grifters, swivel-eyed loons, eugenicists and alleged associators with sex-trafficking pedophiles, forcing everyone to dance to the dreams of these bastards and lunatics innovators and thought leaders, until the bottom falls out of the market, and the only people who stay are the ones who were doing proper work in the first place.

So when someone tells me, “AI is here to stay”, I roll my eyes. Yeah, like herpes, mate. When the bubble pops, everyone will insist that what they're doing is Proper Computer Science™ and it's normal technology, until people stop remembering the insanity that they willingly immersed themselves in and the company they kept with cod-authoritarian wankers and people doing the equivalent of “Tomorrow Belongs to Me” to tell everyone that this technology is inevitable and cannot be stopped.

Come on. We should know the game already. It's been happening over and over again for 70 years.

[Pascal Monett]

Well put

Right down to the last sentence.

The basic problem is that the marketers driving the hype are young and haven't a clue about anything technical, so whatever has happened in the last 70 years is beyond their comprehension.

Hey, as far as tech is concerned, the only thing they know is how to calculate their monthly bonus.

That's why hard disks went from being labelled 1GB = 1024MB to 1GB = 1000MB.

Yeah, I was there when it was forced upon us. The reason that was touted was that it was easier for "the public" to understand.

Bullshit.

"The public" doesn't know what a megabyte is, much less a gigabyte. And don't even go near terabytes, that's lunar tech for "the public".

No, it was just a flimsy excuse for marketing to have an easier time labelling things they can start counting, instead for bothering with actual computing references.

[Elongated Muskrat]

Re: Well put

I can't remember a time when a 1GB hard disk was ever anything other than 1,000 MB; a 1GiB HD, on the other hand would be 1,024 MiB. All that happened was that the ambiguous terminology got sorted out.

[Arkitekt]

Hasnt changed my life. And I haven't used it once. Im good, have no use for it, and think its one of the dumbest things mankind has ever conceived.

Get used to THAT.

[werdsmith]

This will come as a big surprise,l to you, but it’s not all about you.

[drankinatty]

The 10.6 Millon Worker Dirty Little Secret

Well, I think you pretty much nailed it in your assessment. While there is only so much room in one article, one of the economic issues generally skimmed over is a peek behind the 10.6 million jobs that go away.

Raising $30B/Q and promising a Genie that grans wishes isn't a business strategy, it smells more like investor fraud with VCs scrambling like rats on a sinking ship praying an IPO happens soon or many will learn rats don't float.

PEs at 100+ are nothing but hype driven insanity masquerading as finance while regulatory capture allows the charade to continue unabated. The rats deserve what they get, but when the ship slips under, there will be scores of passive investors on their way to the bottom with it.

Whether the current crop of companies pull a rabbit out of the hat and succeed, or are more likely seen playing the part of the ship, the job losses will be severe. And, behind a majority of each of those jobs is a mortgage, and or auto loan(s) and consumer debt the wages serviced, on top of the medical benefits and purpose the job provided. Whether a hater, promoter or seasoned soul waiting to see how the chips fall, we have to recon with the fact that a handful of unelected tech bros are sealing the fate of 10.6 million plus of our neighbors, friends and family who never had a say in the matter. We all pay the cost for the secondary foreclosures and defaults.

Capitalism is one thing, but predatory capitalism is quite another and unfortunately, we will all likely pay the price for the latter for years to come.

[Pascal Monett]

Re: The 10.6 Millon Worker Dirty Little Secret

Capitalism is like AI : it's fine when it is controlled and guided.

The myth of the auto-correcting market is drowned out by the ultra-billionaires who amply demonstrate their ability to shape the market, either economically or with well-maintained relationships with "people in power".

There are solutions to this conundrum, but none that can be applied as long as "the people in power" are ready to sell their citizens into slavery in exchange for lavish parties - including on secluded islands.

So the first problem to solve is : how do we ensure that "the people in power" do their fucking job, instead of grifting everything they can ?

Moral education is the only answer, and that has quite obviously failed us.

[Electronics'R'Us]

I am also old enough

I'm old enough to remember warnings that calculators would destroy our ability to do math

See title.

This isn't my first rodeo with this either.

There was a fascinating study done in the early 70s (I need to find it - must be available somewhere) on the effects of calculators by a UK university.

There two groups; control and test. Participants had a calculator but the test group calculators had been rigged to give implausibly wrong answers (such as 7 * 8 giving an answer of a few million) and all had a multiple choice answer sheet (which included said implausible answers as well as the correct answers)

A sizeable majority of the test group ticked the implausible answer. This is the same effect of all automation where the answer(s) given are relied on because the 'computer says so' and shows just why this issue is not new at all.

Don't get me wrong - calculators are great, but I still do quite a few calculations by hand as I can look at a circuit and have an idea of what the results should be close to. Likewise with waveguide - I can eyeball it and have a good idea of what frequency range it was designed for.

So, yes - automation of all types can lead to mental atrophy and the only way to avoid it is to be able to do what the automation does even if it is at a much slower pace (I am aware that the flight controls for the B2 are far better at maintaining stability than any human ever could be and in fact are the only reason that aircraft can fly safely).

So the apparent argument here is a fallacy - we know for a fact that blindly using any automation tool (be it a calculator or something else) does cause atrophy if we don't exercise our own grey matter as well.

AF447 went down in mid-Atlantic at least in part because the crew at the controls did not know how to fly; they had completely relied on the automation.

[Anonymous Coward]

I thank the Flying Spaghetti Monster every day

I’m not part of the AI or Social Media cults.

[MazeFrame]

Citation needed

While the genie never in history went back into the bottle, this hype cycle is coming of the rails right now.

Computer assisted pattern recognition has proved useful for sorting trash, checking medical images or assisting drivers. Those applications have existed for a long while, can fit into a power budget of 200W on current hardware and have obvious uses in daily life or professions.

Computers were allowed to be non-deterministic when Babbage, Turing, Zuse, Newman, Neumann, Hopper and countless others did their pioneering work. Relays getting stuck and tubes burning out used to be okay. Then computers went to space and landed on peoples desks. Since then, getting computing wrong largely became unacceptable. Recent prominent entries are Intel Pentium FDIV bug or AMD Ryzen segfault.

And then again starting in 2022 when silicon valley lost their collective marbles over text synthesis.

Since the current iteration of LLMs is for all practical purposes non-deterministic, and as IBM noted in 1979, tools can not be held accountable, what do we have for our 1600 billion dollars?

Novelty search? Questionable auto-complete? Eh...?

AI-boosters should be banned from speaking in future tense here!

"Research house Forrester, for example, predicts that by 2030 AI will wipe out 10.4 million US jobs. That's 6.1 percent of all US jobs if you're counting at home."

Taking US BLS numbers, US unemployment depending on time frame, is adjusted for season, up by 0.8% since 2019. Extrapolating from that, good luck. As for ten million out of a job, that would mean 2009 or COVID levels of unemployment. Should ring alarm bells round the globe, if true. Since it is mostly quiet outside of the chronically online folk and some rushed legislation based on hype, TBD here.

"Eventually, though, AI will be cost-effective and efficient enough to kill off many white-collar jobs. It just won't be as soon as many bosses want."

Based on what? The trend presented by the numbers so far shows a linear improvement for a exponential increase in compute. That is non-ideal if the goal ever was to turn a profit. Throwing more data is impossible (outside of Fahrenheit 451-style approach to books). Either someone invents new math or this frontier is a brick wall.

"There are a plethora of AI businesses, but I suspect many of them are dead companies walking."

Are there though?

OpenAI, Anthropic, and...? Who? Anyone else?

"Another issue is that, thanks to AI, we're increasingly unable to tell the truth from fiction."

That was the case before. Education was axed and simple answeres preferred over the correct ones. People are either ignorant to their world or too dumb to notice the consequences to their actions.

"AI has gotten good at coding, but it still sucks at facts, and I don't see that changing anytime soon. "

Has it? Has anyone done research on this? Take some models, take some devs, take some regular folk. Same task, compare efficiency of the solution, in Big-O notation, CPU-cycles or watt-hours. Any citation?

"A decade from now, we'll all be using AI, and our kids will be wondering what all the fuss was about"

They will learn about it in school, probably in a politics or economy class, in a lineup of 1929, (2000), 2008 and 2028. Alternatively try Schumpeter:Capitalism, Socialism & Democracy, Chapter 8 to 10. I'll wait.

"proclaim yourself a 21st-century Luddite"

I will. Because they were right. Working conditions in the factories were atrocious to deadly. The society that developed from there, the have's, the have lot's and the poor sods is ongoing. Outcome TBD.

As with the calculator example, the human mind is a serious case of "use it or lose it".

[Elongated Muskrat]

Re: Citation needed

"AI has gotten good at coding, but it still sucks at facts, and I don't see that changing anytime soon. "

Has it? Has anyone done research on this? Take some models, take some devs, take some regular folk. Same task, compare efficiency of the solution, in Big-O notation, CPU-cycles or watt-hours. Any citation?

We've done some small-scale testing (yes I know, data is not the plural of anecdote, etc.), and it has not got good at coding. Not for anything non-trivial. Maybe for yet another vibe-coded trash app that nobody wanted, but can you see the issue with that?

[Excused Boots]

My, now late, father used to work at the Ford plant at Dagenham in the UK. He once told me a story he had been told by ‘old hands’ at the plant, ‘a manager had proudly shown off the new robots on the production line to a shop steward (union rep); saying how these will never go on strike. True, replied the union rep, but of course they will never ever purchase a Ford car either’.

Now, I have no doubt that this is completely apocryphal and versions have been reported for ages, but the point is valid. No AI agent or LLM is ever going to purchase whatever it is you are selling. So when large numbers of people are out of work, replaced by LLMs, just who are you going to sell your products to?

Yes, yes I know, modern corporate culture is such that it can’t possibly see beyond next quarter's financial results - it’s all short-termism.

But, maybe, let’s just stop and look at this, OK, fine, AI* might be here to stay, how do we stop it having short-term really, really bad consequences?

* Disclaimer, it’s not actually intelligent, well not as humans would classify it. I once heard AI/LLMs as being described as a ‘really complex spellchecker and auto-complete engine’.

[doublelayer]

To the extent that those using LLMs are thinking about that question, their answer is that people working at other companies that LLMs couldn't replace will still have jobs and will buy their products. Most LLM fans don't think that they will eliminate all the workers, just a subset. In reality, LLMs will be unlikely to successfully automate those, and the companies that adopt them and fire most of their employees will fail, not because there's nobody around who can buy their products, but because there's nobody around who wants to buy their products.

To the extent that people predict that LLMs will successfully replace all or at least most labor, they also tend to propose something like a universal basic income to deal with that problem. These include AI company leaders who want to make their products sound more capable than they are and people who already want a UBI and are hoping to use this as justification. In other words, both of the most prominent groups who express this prediction don't really believe it and have ulterior motives for making the claim. We're going to have lots of economic problems from LLMs, but not that.

[phogan99]

We maybe still using AI, indeed we've been using some form of it and machine learning for awhile. I highly doubt the unsustainable debt financed data center boom is long for this world. Nor is it likely to be the hyper growth industry Silicon Valley wants when the bubble finally pops.

The reason people hate AI is entirely the fault of the companies pushing it the hardest.

[MGJ]

Emperors New Clothes

Were similar articles not being written 10 years ago about VR, ie Second Life etc transforming the worlds of work and leisure? How much did Meta write off from their investment in the Metaverse?

[Anonymous Coward]

Re: Emperors New Clothes

Somewhere between 75 and 85 billion dollars over several years, although I don't think they ever released a breakdown of this figure.

Now there are a number of companies each spending that amount every 3 months on supporting LLMs.

[Mohsin2]

I think there’s also a hardware side to this that gets overlooked. Even if AI makes some parts of software work faster, people still need decent machines to actually use these tools. I was looking at the laptop Z11 from Viper recently and it got me thinking about how much the hardware itself matters when your daily workflow is increasingly AI based.

Still, I agree that the bigger issue is the jobs side. Making people more productive is one thing, but removing the entry level work where people normally learn the trade is a different problem altogether.

[Tim Chuma]

The Register posts too many stories about AI

Have read complaints that people are sick of your website posting too much about it.

The only people who actually care about AI are venture capitlists who want to have a pull over their money and people who are being ripped off. Sure, steal 100Gb+ off my website and bust my data cap for the website for the month to train your stupid AI!

[b1k3rdude]

Regarding the sentence - "First, it's going to hit junior employees, like entry-level programmers. Who wants to pay money to someone fresh out of college to do work that Claude Code, OpenAI Codex, or GitHub Copilot can do? No one,"

Er no. Firstly its already been shown that ML generated code is $lop and secondly the cost of the tokens have ALREADY exceeded the cost of the junior engineer, rendering the whole fcuking ML/LLM problem looking for a solution situation dead on arrival.

Following on from what @Tim Chuma said, find other topics to talk about. The majority of your readers have zero fcuks to give about anything Ai related, and why would they when all anyone has to do is look at the STATE of the world economy.

[Jonjonz]

AI accelerates the replacement of the general prosperity that came out of the post WWII economic boom with the K economy. As Mark Twain once said "You tell me whar a man gits his corn pone, en I’ll tell you what his ’pinions is." Thanks to AI being really good at enabling the concentration of wealth, centralization, and making everyone including governments dependent on it, the lucky few in the top of the K economy think everything is just fine, if you're in the lower part of the K economy, not so much.

[doublelayer]

"the lucky few in the top of the K economy think everything is just fine, if you're in the lower part of the K economy, not so much."

It's less about whether everything is fine and more about whether there's a way to make it better. I can name a bunch of things that aren't fine, but that doesn't change whether it's easy to fix them. This argument is often used (though I don't know that you are) as a false dichotomy suggesting that, if we admit there's a problem, it justifies whatever proposal the person who says it isn't fine wants to do. If that proposal wouldn't fix the problem, or if it would cause a lot of chaos in the process, then it probably should be rejected even though the problem is real while we try to come up with a good enough solution.

[imanidiot]

House of cards

The entire premise of this article, and AI as a whole is build on this one premise:

"AI will still be standing and fundamentally change our lives and work"

Hard disagree.

[Anonymous Coward]

The "comforting lie" is that those leading the AI push, the hyperscalars, the AI datacentre builders, the investors, have a plan and know what they're doing, and that the business media would be quick to point out to their readers where the numbers don't add up.

There seems to be a perception that when it all collapses financially it will leave behind lots of useful hardware (as for dotcom boom) and there will be a government (i.e. taxpayer) bail out. Neither of these will happen, the systems in those AI datacentres that have actually gone live have no other use and governments will not be able to cover the losses without destabilising their currencies.

Best case scenario is that the collapse is triggered by customers switching to free/local models, minimising the direct impact on their business.

[Andrew Williams]

Here to stay...

Probably, but the so are serial killers.

Although, AI seems to be getting pricier by the day. Maybe it will for most people just cost too much.

[SparkE]

Poor impulse control

Im so tired of this defeatist attitude of “Genie is out of the bottle, Ai is here to stay” capitulation.

You know what else is out of the bottle, smallpox, polio, ebola, etc etc and we don’t just throw up our hands and say “Oh, well, those diseases are here to stay”

No, we develop ways to stop things that harm humankind.

Many of you want to play with fire and not get burned.

“Oh, Ai is bad so I only use Ai for X, never for Y”…. Yeah, just like getting “a little pregnant”. History is filled with cautionary tales about things too good to be true and otherwise smart, yet willfully ignorant humans continue to blindly ignore the lessons from our past. “Pandora, anyone? Bloody hubris IMO.

Im going to make it really clear, Ai is the new crack epidemic created by the rich. People just can’t resist it and of course, the first hits are free. Once a sufficient number of people become dependent on Ai, the flow of free information is over and you are all going to be groomed to accept whatever the Billionaires in charge want you to eat, think and do.

[Anonymous Coward]

“Eventually, though, AI will be cost-effective and efficient enough to kill off many white-collar jobs.“

That’s quite the assumption. I’ve yet to see any evidence that backs it up.