Posts by LionelB

OpenAI says its latest model is less likely to beat around the bush

Re: I always wondered if

> You know, like real humans do.

Do they? Hung out with humans lately?

Don't believe the hyperscalers! AI can't cure the climate crisis

> There's been reputable climate scientists saying exactly that for decades.

Would you care to share some sources (peer-reviewed)?

> I didn't suggest killing folks off. Nobody did. Strange argument to propose.

Of course I wasn't suggesting that you (or in fact anyone else) were suggesting that! You misread me. Just making the rather obvious point that getting population numbers down means not just lowering birthrates, but also waiting around for people to die off. And since there are an awful lot of young people in the world, that's.. well, large fractions of a lifetime in the future.

> That's obviously wrong.

I don't think so; see above. Yes, population does begin to decrease, but very slowly. And there's a further lag in reduced population numbers having an impact on resource usage and energy consumption.

Don't get me wrong; yes, I think there are too many people on the planet. But we cannot afford to simply sit around and wait for birthrates to fall incrementally with minimal impact on resource demands.

> Yes, but that will take another hundred years or more. Its not a this year thing, where job automation and birth rate reduction could be.

So how do you reduce birthrates, short of imposing limits (as the Chinese tried)? That is socially and politically frought, to say the least. (And no, education doesn't necessarily take "a hundred years or more"; in particular, educated families have less children on average.)

> No, that's a more idealistically achievable idea, but its certainly not now nor ever will be realistic.

Idealistic maybe, but do you have anything approaching a realistic plan to reduce birthrates rapidly to the point where they'll have any impact on resource consumption(see above)? I'm not seeing it.

And that's a problem. People go "Oh, but, global population size…" (always, of course, without realistic proposals on how to reduce it rapidly) with the unstated subtext that we needn't bother trying anything else until that's sorted out, so it's business as usual. In other words, it becomes a defeatist distraction from things we actually can do in terms of, e.g., sustainable energy generation, etc.

Re: still at it with the man-made climate change scam?

> man-made CO2 is PLANT FOOD

Errm, since atmospheric CO2 levels are rising, doesn't that mean that plants are not eating all that yummy plant food fast enough? Maybe we ought to think about not chopping down all those rainforests, and destroying habitats?

Oh, but wait… climate warming induced by man-made rising CO2 levels actually exacerbates destruction of habitats, via various factors such as acidification and warming of the oceans, rising sea levels, extreme weather events, wildfires, flooding and erosion, desertification, …

Duh.

You are not a climate scientist. You are not a biologist. You are not an ecologist. You are a poster-child for the Dunning-Kruger effect.

> The problem is, some are saying it is already too late, and many are warning that we'll reach the precipice in a decade [my emphasis]

But which "some" and "many" were those? Not reputable climate scientists.

> The real problem is that the same "many" were saying the exact same thing 20 or even 30 years ago.

Peer-reviewed science, however, was not saying those things. The problem is the science is frequently drowned out by noisy agenda-pushers (on both sides).

Also, regarding birth rates: the problem here is, that you can't just kill people off. With the best will in the world (clearly lacking), to realistically bring the global population size down to a point where it will make even a jot of difference to energy consumption and greenhouse gas emissions is likely to be on the order of 5-10 decades. FWIW, the best proven way to reduce birth rates is to educate people.

A more realistically achievable goal is to address the extreme skew in resource consumption.

OpenAI’s Altman says Pentagon set ‘scary precedent’ binning Anthropic

I thought it was bleeding obvious, but simply this: companies may choose whom they do business with ­and; history will judge them on their choices.

> If OpenAI or Anthropic want to shut themselves out of that lucrative market, it's their choice I guess.

Zyklon B ring any bells?

AI models suck slightly less at math than they did last year

Re: This is ridiculous

Seems about right to me – most humans suck at maths. That's why we invented calculators and actual maths software.

(Clue: we're talking about Large Language Models, trained on human data; why on earth would they be good at maths?)

Meh. Wrong tool for the job. That's not what they were designed for.

Re: If I bought a two-euro calculator

Damn right. I asked an AI what the age of the universe was, and it told me it was 13.7 billion years. I immediately asked it again, and it had added 7 seconds to its previous answer!! Then I asked it twice to give me a number between 1 and 10 and again got two completely different answers.

Bloody AI, eh?

Re: Easy fix

Even easier solution. Use a bloody calculator. Or if it's a complex problem, use a maths package like Mathematica.

Why on earth would anyone use an LLM to do maths? They're large language models for chrissake, obviously wrong tool for the job.

This is just classic PEBCAK.

UK tech hit by double trouble: Fewer foreign techies amid skills squeeze

Academic salaries (especially at postdoc/research level, as opposed to faculty teaching posts) are not great in either the US or UK. Depends a lot on the funding body.

The most egregious despots in history, past and present, had not "had a whirl in power" – until they did. I'm not sure "advantage" is the word you're looking for.

> Who'd want to come to the UK to work with the rise of Farage and his fascist cronies?

Americans, that's who. The orange fascist and his cronies have already arisen over there, and hold the reins.

Don't know so much about the tech sector, but we're certainly seeing that in academia; in the US academia is under siege, and funding is drying up. The number of US applicants for UK and EU funding has surged of late (I believe Canada is seeing something similar). I suppose we could be beneficiaries of the US brain-drain, but it does pile pressure on home-grown applicants.

Google presses play on 30-second Gemini musical slop generator

Sure. And opera is for people who think singing is bellowing like a moose in mating season (men) or creating high-volume sine waves to shatter nearby glass objects (women). Western classical music is for those who are challenged by syncopation or rhythmic complexity. Heavy metal is for those who enjoy gurning and the amplified sounds of squalling infants, and believe that more notes-per-second on a guitar is better (doesn't matter which notes). Jazz is for people who don't really like music but want others to think that they do. Pop music is for singers who can't rap.

Any more random prejudices we can throw in there?

Of course. And you can say exactly the same about any genre of music since the first proto-humans beat two sticks together.

What was your point again?

You've simply revealed that you don't "get" punk¹.

That's fine; I don't get heavy metal. Or opera (for me, the theatrical aspect is risible, and the singing style grates my brain). But I'm not going to write those genres off, as those are entirely personal takes, obviously not shared by millions, and (unlike yourself) do not consider myself an arbiter of musical taste.

AI-generated music is, to my ears, entirely lacking in talent and inspiration – but more cogently, it lacks soul².

¹_{Of course, as with any music genre, there was plenty of [human-generated] punk slop back in the day. It's also a misconception that the purveyors of (good) punk had no talent, and plain wrong to say they had no inspiration. You don't get to change the trajectory of popular music without those things.}

²_{I had a listen to that "afrobeats" clip (as it happens, I play African music myself). It is godawful, a pale, white approximation to the real thing. It sounds like the soundrack to a cliched ad for a fruit-based soft drink slapped together by a bunch of talentless, uninspired, soulless suits. It does to afrobeats what Peter Andre did to reggae back in the day (which might be described as f****** it up the a***).}

Gemini users say their chat histories have quietly vanished

Re: Mine's still there...

Actually, as a mathematician/statistician, it's come through pretty well for me; I use it primarily to help me get up to speed with techniques and references in areas which are outside my immediate expertise. Doesn't always get it right (although it does more often than not), but that's okay, as I'm going to follow up the references anyway. (And no, I've not had issues with hallucinated references.) In short, it's a genuine time-saver.

I don't use it for coding or actually solving mathematical problems.

Depends what you're using it for, I guess; YMMV.

Google germinates Gemini 3.1 Pro in ongoing AI model race

Re: Doubletalk as usual

> That is precisely what humans do when driving, and in many other situations.

Well, we try to. Essentially, we make informed guesses (informed, as you say, by empathy – "You're like me" – among other things). I.e., we use a probabilistic approach, which has been my contention all along.

> That is why we are so much better than self driving cars at predicting the behaviour of cyclists (we can imagine when they will behave like vehicles and when they will behave like pedestrians) …

Yes. In principle, a successful self-driving vehicle will need to model at least basic human psychology – to "empathise" with humans, if you like. Maybe there are already attempts along those lines, but that's a massive ask.

Ironically, if all road users were AI, that would be not so much a problem (as long as they all had access to each others' algorithms).

> … and why autistic people have so much trouble understanding why other people behave as they do.

Yes.

> "AI" has no theory of mind or concept of anything. All it can go by is what people have done and not why they did it.

Agreed. Although people are not necessarily that great at understanding why they, or others, do the things they do. Our "theory of mind" is rather poor. We do, however, have advantages here called "communication" and "culture" – we interact with each other, while literature, film, theatre, … also help us gain insight into how people behave. Purely speculative, but perhaps we will not have (human-like) machine intelligence until AIs communicate with each other (and us ­– that already happens) and develop some form of "culture".

We also, of course, have the ability to introspect, which is a great advantage when it comes to empathy. That is not in principle unimaginable for machine intelligence, and I believe there is/has been research along those lines. (Machine introspection, though, is not going to yield insights into human psychology!)

Re: Doubletalk as usual

Not sure why the downvotes here. This seems rather uncontentious to me. Someone care to comment?

Our brains are literally (noisy) embodied neural networks, which are ultimately the product of billions of years of evolution, and whose functionality is based on that evolved structure plus lifetime learning.

There is, furthermore, good evidence from cognitive neuroscience that our cognitive functionality (including those "heuristics") involves statistical processing. This stands to reason; in an unpredictable world, statistics is frequently as good as it gets. In particular, brains are not digital computer-like deterministic machines cranking away through digital computer-like algorithms. (That may even be one reason—apart from sheer scale as well as our poor understanding of human cognitive processes—why it is so hard to implement human-like intelligence on digital computers.)

Re: Doubletalk as usual

Indeed not :-)

> Form follows function. If you wish AI to ape human intelligence, then of course it MUST use human processes.

The problem is we have a very poor understanding of those processes (and I say this as someone working in a cognitive neuroscience-adjacent, rather than machine-learning field, although there is some overlap these days).

Re: Doubletalk as usual

Sure we do – those will have been "programmed" by evolution and lifetime learning. And those heuristics may even have a statistical element.

Re: Doubletalk as usual

Has it occurred to you that human intelligence involves rather a lot of 'Analysed Probabilities'? When we navigate our way through a complex world, that's pretty much exactly what we do. In fact it's pretty much all you can do when your basis for predicting what's going to happen next, and how your own actions will affect that, is severely constrained by the information available to you. When you're driving a car, for example, you're constantly predicting what other drivers are going to do, and what's around the next bend. But you can't get into other drivers' heads, and you can't see around corners; so you're effectively weighing up probabilities in real time.

As for "reasoning", interestingly the big models seem to be moving away from pure LLMs towards hybrid models that deploy various learning mechanisms and logical manipulation. Ironically, this echoes early (failed) attempts at AI, in the 60s, 70s and 80s, when researchers imagined you could "reason" your way through complex situations via pure formal (or at least fuzzy) logic. That turned out to be a blind alley – those attempts essentially foundered on combinatorial explosions; hence the subsequent move towards statistical network-based models. Bear in mind (literally!) that humans intelligence is underpinned by noisy neural networks, not formal logic gates. Our intelligence is more statistically grounded than you might imagine.

Of course I am not claiming that current AI is anywhere near human levels of intelligence, and deplore the hype as much as the next person. (Then again, it's hardly a level playing field, given our several billion years' worth of evolutionary R&D and mad hardwetware advantage in terms of processing power.)

UK to demand social platforms take down abusive intimate images within 48 hours

No, I'm from the UK; but I was thinking more of the US, where the law seems to have no appetite to take on the tech giants (one might well ask why that is the case).

But even in the UK and the EU, legal sanctions against the big players seem to amount to a slap on the wrist. Hopefully this will change.

As for the posters, if they're using VPNs or Tor, identifying them (and the jurisdiction they're in) is hard to impossible even for the platforms and ISPs.

Agreed in priciple, but surely the problem lies with the "cooperation of the platform/ISP" part. That doesn't happen unless you legislate for the mandatory cooperation of the platform/ISP which, as we know, runs into all kinds of problems.

> I can't help feeling the message to the public is correct but the obvious tentacle of Government censorship of entire foreign websites/domains rather than dealing with illegal posters is driving this.

I don't personally feel that ordering the takedown of illegal, abusive content targetted at individuals amounts to censorship – YMMV.

And yes, it would be lovely if we could go directly after the illegal content posters rather than the platforms, but how do you propose that might be achieved (without far more intrusive state/legislative intervention)?

Linus Torvalds and friends tell The Reg how Linux solo act became a global jam session

Re: It's a remarkable thing

I suggest that you have a massive chip on your shoulder.

Re: It's a remarkable thing

> I can't think of anything else to thank him for - anyone else have any other suggestions?

Um, how about for creating Linux and overseeing its development and maintenance over the past 35+ years?

Price of popularity: Linux Mint's success also means maintainer stress

Mint user – I basically live in the terminal.

There seems to be a curious misapprehension that Mint is exclusively a "newbie" distro. Really not sure why that is. I'm a Linux user for almost three decades, and use it for scientific computing and software development. I choose Mint these days because of the ease of installation and configuration, and useful tools. If I need to do anything funky at a system level (which is rare) Mint does not get in my way.

> It's always the user's fault in FOSSland, isn't it? But never, oddly enough, when Windows cocks up. That's always Microsoft's fault.

Wind your neck in, I wasn't having a dig at you personally. When someone reports a screed of problems—with any system—which do not seem to plague others in a major way, it seems reasonable to suspect hardware issues or a broken system (which may or may not be down to the user).

Okay, so CachyOS sounds interesting (if hard work), but, like most distros it supports a range of desktops. Are you perhaps actually saying that you find Xfce "boring"? Which desktop do you run on CachyOS? And why not try a different desktop (on Mint)?

_{As it happens, I don't generally like or need a full desktop environment; I usually run the minimalist Fluxbox window manager—on Mint—without a desktop. I'd say that Fluxbox has a rather distinctive "personality".}

My car is really boring too. Always goes in the direction I point the steering wheel, accelerates when I push the accelerator, and slows down when I hit the breaks. Clearly I need to find something more exciting …

Sounds to me rather like you have hardware problems, and/or somehow managed to break your system(s).

Mint with Xfce has always, aside from a couple of minor and fixable issues, pretty much "just worked" for me (including sound, scanning, printing and LibreOffice) – for many years, and on a range of desktops and laptops.

Ireland joins regulator smackdown after X's Grok AI accused of undressing people

I'm going to suggest that if the undresee were a child, and the images/videos freely and publicly available, their parents might beg to differ with you (and that's just the most extreme end of the issue).

FFS, I cannot see even the remotes justification for the facility to create and disseminate such images to be readily available to all and sundry — "freedom of expression" does not even come close. Surely even the most zealous Musk fanboi would concede that freedom of expression is not unconstrained (unless you're a very literal anarchist). I cannot, for example, express my freedom to arbitrarily kill someone I don't like, because, y'know, it's against the law. If laws are currently inadequate to protect people against a novel form of obvious and clearcut abuse, and provide no redress against such abuse, then the law needs to catch up.

Re: a cynical attempt to shutdown the last truly open forum on the Internet

Dear AC. Just a heads-up that your post could be interpreted as a cynical attempt to promote the creation and dissemination of pornographic images of children.

Not that I'd interpret it that way myself.

Oh, no, definitely not.

GPT-5 bests human judges in legal smack down

Re: AC so the bots don't get me

Yes, that is a legitimate concern – hence the importance of human oversight.

I'd like to think that might be less true for judges – you'd have thought laziness would be an immediate (and rather easy to spot) disqualification for the role. But hey ho, …

Yes. In England there is an independent Judicial Appointments Commission (and a separate independent commission for Supreme Court appointments).

No, that wasn't the argument. Did you read the article through to the end, or just the (sensationalist) byline?

"As opposed to a Republican … or a Democrat judge's hallucinationprejudice or political allegiance?"

At least in the UK judges are not (yet, or overtly) political appointments.

I don't see that happening any time soon. I simply don't see judges buying it, and it would surely entail some massive (and massively contentious) legislation to push that through.

A more likely scenario is that LLMs are used to check formal legal arguments, precedents, etc. (which according to the article they may be rather good at). That would then be reviewed by the (human) judges, at which point any hallucinations would be revealed. That is, AI used as a tool, but with human oversight.

> And how is AI to know if a witness is reliable or telling complete porkies?

To be fair, humans (with no exception for judges and juries) are not terribly good at that.

Elon Musk paints exodus of xAI co-founders as 'evolution'

Re: He isn't completely wrong

> It doesn't work that way anymore.

It never much did, really. See my previous post.

> Harsh, but the culls are afforded special care to ensure their survival.

Are you referring to conservation of endangered species by humans? Well, if species are endangered these days, it's most likely to be by us. Attempts at conservation are overwhelmed by the current catastrophic loss of species by human destruction and degradation of environments.

Unless you meant something else?

Re: He isn't completely wrong

That is known as the saltation theory of evolution, and is largely discredited (although it's more nuanced than that… let's just say it's rare – the Wikipedia article gives a lot of detail).

As for "in-betweens" in the fossil record, in fact we see many. It may seem that we don't, because evolution does not proceed at a steady pace; so, for example, in some set of species there may be comparatively little major evolutionary change over, say, 10,000,000 years, followed by a "rapid" change over a mere, say, 1,000,000 years, followed by another 10,000,000 years of comparative stasis. During that rapid period not so many fossils are layed down, so those "transitional" fossils are rare. This dynamic of (Darwinian) evolution is sometimes known as punctuated equilibrium, and may be accentuated by the fact that the laying down of fossil-bearing layers over geological time is also patchy.

Re: He isn't completely wrong

Pedantic points:

b) Not all changes are for the better - the majority are not significant or are terminal.

You're thinking of mutation there – one of the drivers of evolution, not the thing itself.

c) The value of any improvement is rarely immediate - most are incremental improvements that are only realised after several generations.

That may or not be the case – although improvements are indeed most often incremental. More accurate to say that the (rare) advantageous mutation will generally take many, many generations to "fixate" in a population; i.e., to become the norm. (This depends, among other things, on the magnitude of advantage conferred, and on the size of the population.)

The big FOSS vendors don't eat their own dogfood – they pay for proprietary groupware

Re: Too much ideology makes Liam unproductive

The article, you mean?

Re: Sales argument

Then again, who'd want to argue that that's worked for Google or Microsoft?

OpenAI introduces ads...for the people!

You mean like just about every streaming service on the planet?

Edinburgh councillors pull the plug on 'green' AI datacenter

Re: NIMBYs

Errm… so it's not in anyone's back yard. So there cannot by definition be any NIMBYs. You seem confused.

Oh, right, sorry, I'm not thinking like a councillorCity planner.

The Linux mid-life crisis that's an opportunity for Tux-led transformation

Re: Capitalism at its finest

> Sure, all those musicians were happy to live that way and wouldn't have changed it if they could... all of them were looking for success - which means fame and money also. They didn'y look for dying poor.

Perhaps you missed my earlier "That's just silly. I am not claiming any artist would not want to monetise their art "

The rest is just repetitive strawman flam that I wasn't disputing in the first place.