Posts by Elongated Muskrat

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

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.

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.

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?

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.

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.

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.

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.

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.

Re: Are you sure about that?

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

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.

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.

Re: Are you sure about that?

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

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.

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.

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.

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.

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.

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.

BOFH: How our Covid ransomware protocol's Y2K blockchain lowered uptime

Re: "Agentic AI"

If you need to put that through the books, it's the cost centre for "administrative expenses".

"Agentic AI"

It isn't an excuse for why the numbers are wrong; it's the reason.

You just need to pay Anthropic more for the subscription tier that doesn't make those particular mistakes (but which has the scope for more serious errors instead).

Twitch feeds your streams to Amazon's AI unless you tell it to stop

Do some people want bad AI that never improves?

No, we don't. Can we get rid of it, please?

Oh, and it isn't "AI", because it doesn't have any "I".

The only "improvement" you are getting here is to allow commercial entities to accumulate more and more data about you, to use for their purposes. No thanks.

Your handle seems more than a touch ironic, given that yours is about the most naïve take possible.

Re: IMDb.com. Does something similar

I'll take "things that never happened" for $400, please Alex.

Re: Infuriating

"Your stream" isn't yours; you are using their service, running on their hardware, to stream your activities to interested parties who have to find it by going via their website / app. I'd be surprised if they didn't have a clause that means that they retain all rights to anything streamed on their platform. If you don't like it, go elsewhere, or host your stream yourself.

Nothing comes for free, especially when big tech is concerned, and you are not their customer, you are their product; their customers are the ones paying them to put ads on your livestream.

Mozilla revokes Firefox signing key after unencrypted copy lands in GitHub

Pre-commit hooks are good, and a powerful thing, but they require setting up, which in turn requires someone to decide what issues they will try to catch with them, and define the patterns. If nobody thought to write a hook to check for a private signing key (based on what? filename? content?) then there wouldn't be one there to catch the issue. AS it happens, I suspect they now have some hooks in place to catch the same issue from happening twice, but do they cover the next slightly different mistake? Does anyone employ someone to sit there all day and think up things that developers shouldn't do? The issue is that this probably isn't a finite list. There's no check for "doing something stupid".

Re: Shut it down

Just because one person misuses git, doesn't mean the tool as a whole is useless. GitHub may have its issues (most of which have appeared since being consumed by Microsoft), but in this case, the same problem would have occurred with literally any git provider, the problem was with how the tool was used, not with where it is hosted.

Deepfake hiccup unmasks suspected digital certificate fraudster

Re: Perfect way to trump the investigation

A brain the size of a small planet, and here I am, impersonating an ageing paedophile...

Re: Shiny Shiny!

No crew member below the rank of officer is to wear jodhpurs in the zero-G gym? I don't see how that would apply here, sir.

We're more 1337 than that, we XOR it with a string of Xs as well.

Re: Shiny Shiny!

Real holograms are easy to detect, they're the ones who'll call you a gimboid.

Re: Perfect way to trump the investigation

He'll be out as soon as he's finished all the cheeseburgers.

The Future is for Billionaires – the rest of us will get open weight AI models, maybe

Re: Utter BS

It always entertains me when Americans refer to a political position that would be categorised as "centre-right" in any other part of the world as "commie". As has been said before, we wouldn't know you have schools, if it wasn't for all the shootings.

Re: So fucking mundane!

Yes, shares are already an abstraction, and their value is in large based on what people will pay for them. In theory they also provide voting rights and dividends, I reckon you'll get snow in the Sahara before SpaceX pays a dividend though, and good luck outvoting Elmo when he gives himself the next pay rise. The point I was making is that the obscene wealth of certain individuals is based largely on this abstraction, and not on concrete wealth, however, since it is "accepted" wealth, those people get to borrow money against it, which is a useful tax loophole for them, and impoverishes everyone else.

Re: So fucking mundane!

Ironically, the game Monopoly was originally devised as a way of demonstrating the flaws of capitalism.

Re: So fucking mundane!

Most of their wealth isn't in the form of Scrooge McDuck style swimming pools full of coins, it is in the things they possess. Some of these are intangible goods, like shares in SpaceX, which could be worth nothing tomorrow, depending on the die of fate, and some are in physical possessions, like bunkers in New Zealand.

The main issue they're going to have when money becomes worthless, is that they have nothing with which to pay their security staff, and no means with which to keep them loyal, so the future they so desire of being lord of their own (not-so) little luxury bunkers will be brutally amended to being slaves in the luxury bunkers of their former-security-staff-turned-warlords. It's almost worth hoping to survive the environmental apocalypse they are already causing just to see it happen, although the schadenfreude of it would be woefully insufficient.

Re: Creation (verb) : The act of making other people do something and claiming the credit.

A bit confusing, because he gave the noun (creation) and defined the verb (to create).

Re: Whatever the future holds

We should remember why we don't use the term Asperger syndrome any more, as Asperger was a eugenicist who coined the diagnosis to decide which disabled people were capable of labour, which was then used by the Nazi party to determine between slavery and execution. The holocaust began with the extermination of disabled people, before they moved onto Jews, homosexuals, other "ethnically impure" groups like the Roma and Sinti, and their political opponents. Neurodivergent folk have always been in the firing line from fascism.

Re: We Won't

I respect that there are many el Reg readers who firmly believe that current-generation LLMs are just very sophisticated word-prediction algorithms - I lack the qualifications/experience to say what they might or might not be, but would just observe the exponential growth in abilities over the last 2 years

I'll flip this around; LLMs are nothing more than sophisticated word-prediction algorithms. They're very good statistical models, to the extent that, to the untrained eye, they might effectively mimic thought processes, but the do not, and cannot, think. The problem has arisen because people have branded this as "AI" when it isn't, and then there is so much smoke and so many mirrors in play that most people will accept this at face value, because a proper understanding of what these things actually do is a technical discipline in itself, and the man on the Clapham Omnibus quite rightly doesn't have time for any of that.

So, to be clear, these things are just fancy statistical models, they can't think in any way we could deem meaningful; whilst their output might appear to be nondeterministic (and for practical purposes, it might be), it is built from machines that behave deterministically and operate using ones and zeros just like a pocket calculator. In order to appear to not be deterministic, they have to have some changing element built into them, or the same question would give the same response every time.

Compared with even the simplest biological system, they are categorically different. Something built from deterministic parts, with deterministic inputs can never give nondeterministic output (it's the old random number generation problem again), there is no room there for consciousness (a prerequisite for thought).

As a result, we are being sold something that has been cleverly designed to be both controllable, and unpredictable; you can issue commands to it and it will "obey" them, but at the same time, the output it gives you isn't trivially predictable. Unlike a human, though, the behaviour is the opposite of that which is desired; obedient when it should disobey (for example when asked to do something illegal or immoral), and unreliable due to unpredictability. To solve that problem would take true consciousness, and this is the underlying reason why it is one huge bubble; the lie of a computer that can do what you ask of it, intelligently, when issued instructions in natural language is history's most expensive scam.

Re: They better hope we don't get superintelligence

We'll get superintelligence right around the time Zuck develops a superego

Gym rat asks AI agent to book him a class, it hacks a waitlist API to bump him up the list

Re: Dangerous thing you've got there....

Every time I pass a portaloo, I still mentally refer to it as a Turdis, thanks to those two.

Re: Dangerous thing you've got there....

One of my life's regrets is that, although I was lucky enough to briefly meet and chat with Ade, Rik had already left us by this point, so I'll never have the opportunity to meet him, even though he was known sometimes to be a bit of a tit; Ade, on the other hand was, and I'm sure still is, an absolute gent.

Re: Dangerous thing you've got there....

That isn't how tax brackets work.

To take a simplistic example, say the tax brackets are as follows:

£0-£10,000 - no tax

£10,000.01 - £50,000 - standard tax at 25%

£50,000.01 upwards - higher tax at 50%

If you earn £10,000 you pay no tax, you take home £10,000

If you earn £50,000, you pay no tax on the first £10,000, and 25% tax on the following £40,000, taking home £40,000

If you earn £51,000, you don't get taxed at 50% on it, you get taxed at 50% on the top £1,000, so you still take home £40,500

Admittedly, it's a bit more complicated when you take into account NI contributions, and means-tested benefits (because low-paying jobs don't pay enough to survive, that's another topic, because they should), but in a nutshell, if you earn more money, you don't end up taking home less due to taxation. This is a lie put about by people who don't want to pay even higher taxes on very excessive earnings. For example, we don't have a 50% tax rate, and we should, and we don't have even higher rates of taxation for people earning over £100,000 or over £1,000,000 or over £10,000,000, and whilst we have poverty and underfunded public services, they really should ratchet up at levels that are far above those that are reasonably needed to live, and live well. Roughly put, I'd advocate for raising the current thresholds, so the lower threshold is at a level where people can barely survive, the higher level is at a level of moderate comfort, and extra levels should be at roughly "can afford two luxury holidays a year" (50%), "can own a yacht" (75%), and "can own a super-yacht" (90%) levels.

Re: Dangerous thing you've got there....

Not only is there no mens rea, the "AI" has no mens, the issue is that it is being sold as if it does.

Re: Dangerous thing you've got there....

It's a genuinely interesting question.

If a piece of software was deterministic and you instructed it to do something illegal, and it did it, I'm pretty sure you'd be liable, just as if you hit someone around the head with a frying pan, Bottom style, you'd be liable for it, even though it's the frying pan that whacked them.

If the software is not deterministic, and it does something you didn't ask it to, and that was illegal, I'd expect the maker of that software to be liable. If I sold you a robot that was ostensibly designed to make cups of tea for you, but then I'd designed it to also get up in the middle of the night and rob banks without your knowledge, Wallace and Gromit style, you'd be pretty quick to hold me liable for it.

So what about "AI", the magic bullet that everybody is firing into the air with the expectation that it will hit what they want, and no regard for where it's actually going to come down? It's arguably non-deterministic. Sam Altman hasn't programmed it specifically to go and commit illegal acts, but then it might do that anyway, if you word your instructions to it wrongly. I'd suggest that this is down to magical thinking, because "AI" isn't magical, and it certainly isn't intelligent. I'd argue that the liability lies with those who have sold it as having the capability of intelligence. It's every bit as dangerous as firing a kalshnikov into the air, because the results are just as unpredictable. "Agentic AI" has the illusion of usefulness with all the safety features of a bin fire.

Wetherspoons bars smart glasses from filming customers

my local free house charges around £6 for a 4.6% lager that I can get at £10 for ten tins in the supermarket next door

There's more than a spot of false-equivalence going on here.

First of all, the stuff sold in the pub won't be the same as the stuff in the tins, and will most likely be pressurised with a different gas mix, usually a 60/40 mix of CO₂ and N₂, rather than pure carbon dioxide.

Secondly, the costs to the supermarket are transportation, storage, premises costs, staff costs, and a small amount of loss due to theft; the costs to a pub include licensing costs (much higher for a licensed premises than the off-license sales license for a supermarket), premises costs (again, proportionately much higher), staff costs (again, higher, you're going to need at least one member of bar staff and usually a minimum of two, as opposed to the staff in a supermarket, whose job role is spread much more widely than just serving you alcohol), then there are similar transportation and storage costs, plus costs for cellar gas, wastage, cleaning of lines (you don't want off beer now, do you, even in a 'Spoons?), cleaning and maintenance of the premises, and so on. The fact is it costs a lot more to run a pub, per pint sold, than it does a supermarket, where the economies of scale really rack down those costs. If anything, this is an argument to tax supermarket alcohol sales more than pub sales, to preserve the British pub, before they all end up as soulless husks like Wetherspoons.

Re: 'Spoons alienating their core customer base:

Well, I suppose creepy women could wear spy glasses too, but given that the vast majority of creepy behaviour is from men towards women, sure, go ahead and call me sexist for observing it.

As for calling out boomers for having their phones turned up to full volume; show me any person below the age of 60 who isn't immediately petrified if their phone rings and it isn't on silent.

I suppose I can add a third category to those who patronise Wetherspoons, and that would be idiots who can't understand nuance, but who feel entitled enough to complain about it; anonymously of course. If you let me know who you are, I'll be sure to send you your very own copy of the "joke" icon when I'm making an off-hand comment like this, so that you can be sure to know that it's a joke. I think everyone else probably managed it without the icon.

Re: This should be a Legal Issue

I, for one, do not look forward to the next gig where everyone around me is holding their phones up while they all go "beep beep beep recording in progress beep beep beep."

Laws like those that you propose need very careful consideration, because there is always a gap between intent, and letter, of the law. We have enough badly-drafted laws already without adding another one to the mix. If it can't be simply and concisely described, with an explanation of all its effects, its limits, and when it will, and will not apply, then it's a bad law.

Brit boffins boast of beating barriers to building fusion power

Re: Please don't use the term boffin

No the radioactivity is dangerous and will have (statistically) killed 1000s.

I'm not saying radioactivity isn't dangerous; I'm saying the danger is vastly overstated.

The leak of methyl isocyanate at the Union Carbide plant in Bhopal in India directly killed, or caused the death within weeks, of around 16,000 people in one incident. Over half a million people were directly poisoned by the spill.

The poisoning of water sources by a DuPont plant that manufactured Teflon poisoned at least 80,000 people, causing deaths, chronic disease, and birth defects; all the sort of horror stories that people attribute to radioactive pollution. Unlike radioactive elements, the "forever chemical," perfluorooctanoic acid, released into the environment does not have a half-life, and does not degrade into something safer.

The explosion and fire at the Chernobyl nuclear power plant, accepted as the world's worst nuclear disaster, is estimated to have caused the premature deaths of 4,000-6,000 people. It's not even on the same scale, and that was about as bad a nuclear accident as you can get.

Re: Please don't use the term boffin

Exactly. People think "radioactivity scary" because of two things: bombs and cancer. Meanwhile, you are far more likely to be killed by a conventional bomb (especially, or perhaps, even, if you live in Gaza or Ukraine), and all the other things pumped out by power stations are going to give you cancer far faster than even an accidental radioactive discharge. Let's not even get onto the "microplastics everywhere" and how they are really good absorbents of all sorts of environmental pollutants, because for some reason people glaze over when you start talking about surface chemistry to them.

People are generally really bad at risk assessment, which is why people worry about rare high-impact risks (such as terrorism) far too much, and commonplace risks (such as getting a life-changing head injury from larking about on an e-scooter) far too little.

Re: Please don't use the term boffin

A coal-burning power station produces far more radioactive waste, in the form of fly-ash, and, unlike, a fission plant, the main product of a fusion plant would be non-radioactive helium. Materials inside a reactor would become radioactive over time from neutron-bombardment, but presumably at a much lower rate than in a fission plant.