because that phrase doesn’t ever appear in the training data.

Eh but LLMs abstract. It has seen “<animal> have feathers” and “<animal> have fur” quite a lot of times. The problem isn’t that LLMs can’t reason at all, the problem is that they do employ techniques used in proper reasoning, in particular tracking context throughout the text (cross-attention) but lack techniques necessary for the whole thing, instead relying on confabulation to sound convincing regardless of the BS they spout. Suffices to emulate an Etonian but that’s not a high standard.

Lemmy search already is quite excellent… at least here on lemm.ee, we don’t have many communities but tons of users subscribed to probably about everything on the lemmyverse so the servers have it all.

It might be interesting to team up with something like YaCy: Instances could operate as YaCy peers for everything they have. That is, integrate a p2p search protocol into ActivityPub itself so that also smaller instances can find everything. Ordinary YaCy instances, doing mostly web crawling, can in turn use posts here as interesting starting points.

AI image generators don’t “consult” source images to generate an output.

Well, you have an artist breaking things down for an audience understanding neither the technical nor artistic aspect…

Modern AI generators are increasingly good at generating text. They still struggle a bit

I mean… SDXL still struggles a lot. The only thing you can get it to spell reliably is probably “Hooters”. There’s the one or other lora which makes it not suck completely but it’s still nowhere near actually good at generating text, the training just isn’t there. And even with that in place things like signatures are probably going to be gibberish.

While a naive (and cheaper) approach to AI generation doesn’t use layers, there are generators which do use layers,

Unless you start off training by feeding the model 3d data (say, voxels) alongside 2d projections I don’t think it’s ever going to develop a proper understanding of these kinds of things. Or, differently put: Learning object permanence (of sorts, related) is a meta-cognitive abstraction step that just won’t happen with the type of topologies we know how to engineer. It’s probably like 90% on the way towards AGI, so to get a simple topology to understand it we have to spoon-feed it permanence information alongside the (apparent) non-permanence.

It’s quite easy to trick people with untrained eyes… for one, they have no idea what “consistent illumination” and stuff means. And something being off doesn’t mean that an AI made that mistake because humans make mistakes, too – photographs don’t, but the general problem is not just about telling realistic stuff apart but also illustrations. You’re looking specifically for mistakes that AI is likely to make, but humans are practically never going to make. And yes humans get hands wrong all the time.

Here’s a good video about what to look for and what not.

Even without all that messing with stuff too much is bound to clash with protections those kinds of sites have around editorialising. That is, by doing such stuff X says “we’re not actually a pinboard, we’re a newspaper, we’re editorially responsible for what’s on there”, and then prosecution can come along and say “so, your newspaper published an article calling for <crime>, didn’t it? That’s your speech now, not speech of some random user, isn’t it?”.

Honestly if you’re managing kernel and userspace remotely it’s your own fault if you don’t netboot. Or maybe Microsoft’s don’t know what the netboot situation looks like in windows land.

You call it unregulated, but that is the natural trend for when the only acceptable goal is the greater accumulation of wealth.

Nah unregulated is the exact right word and that isn’t the kind of neolib you’re out for. Those would use “free” instead of unregulated, deliberately confusing unregulated markets with the theoretical model of the free market which allocates resources perfectly – if everyone is perfectly rational and acts on perfect information. Which obviously is not the case in the real world because real-world.

There’s a strain of liberalism which is pretty much the cornerstone of Europe’s economical model, also, generally compatible with socdem approaches, and it says precisely that regulation should be used to bring the real-world market closer to that theoretical ideal – they’re of course not going all-out, you’d need to do stuff like outlaw trade secrets to actually do that, have all advertisement done by an equitable and accountable committee and shit. But by and large regulation does take the edge off capitalism. If you want to see actually unregulated capitalism, have a look at Mexican cartels. Rule of thumb: If you see some market failure, regulate it away. Like make producers of cereal pay for the disposal costs of the packaging they use and suddenly they have an interest in making that packaging more sensible, can’t externalise the cost any more.

Defeating capitalism ultimately is another fight altogether, it’s nothing less than defeating greed – as in not the acquisition of things, but getting addicted to the process of acquisition: The trouble isn’t that people want shit the problem is that they aren’t satisfied once they’ve got what they wanted. Humanity is going to take some more time to learn to not do that, culturally, (and before tankies come along nah look at how corrupt all those ML states were and are same problem different coat of paint), in the meantime regulation, rule of law, democracy, even representative democracy, checks and balances, all that stuff, is indeed a good idea.

Won’t last for long because deflation is built into bitcoin and every sane state matches monetary supply to economical output to keep prices stable. El Salvador isn’t doing that anyway, though, otherwise using USD, or getting many tax payments in bitcoin, the thing being about as liquid as asphalt, so it doesn’t really change much.

That’s already the nvidia approach, upscaling runs on the tensor cores.

And no it’s not something magical it’s just matrix math. AI workloads are lots of convolutions on gigantic, low-precision, floating point matrices. Low-precision because neural networks are robust against random perturbation and more rounding is exactly that, random perturbations, there’s no point in spending electricity and heat on high precision if it doesn’t make the output any better.

The kicker? Those tensor cores are less complicated than ordinary GPU cores. For general-purpose hardware and that also includes consumer-grade GPUs it’s way more sensible to make sure the ALUs can deal with 8-bit floats and leave everything else the same. That stuff is going to be standard by the next generation of even potatoes: Every SoC with an included GPU has enough oomph to sensibly run reasonable inference loads. And with “reasonable” I mean actually quite big, as far as I’m aware e.g. firefox’s inbuilt translation runs on the CPU, the models are small enough.

Nvidia OTOH is very much in the market for AI accelerators and figured it could corner the upscaling market and sell another new generation of cards by making their software rely on those cores even though it could run on the other cores. As AMD demonstrated, their stuff also runs on nvidia hardware.

What’s actually special sauce in that area are the RT cores, that is, accelerators for ray casting though BSP trees. That’s indeed specialised hardware but those things are nowhere near fast enough to compute enough rays for even remotely tolerable outputs which is where all that upscaling/denoising comes into play.

Sounds like an expert system then (just judging by the age) which was AI before the whole machine learning craze, in any case you need to take the same kind of care when integrating them into whatever real-world structures there are.

Medicine used them with quite some success problem being they take a long time to develop because humans need to input expert knowledge, and then they get outdated quite quickly.

Back to the system though: 35 questions is not enough for these kinds of questions. And that’s not an issue of number of questions, but things like body language and tone of voice not being included.

so it’s probably just some points assigned for the answers and maybe some simple arithmetic.

Why yes, that’s all that machine learning is, a bunch of statistics :)

The way to use these kinds of systems is to have the judge came to an independent decision, then, after that’s keyed in, the AI spits out theirs and whichever predicts more danger is then acted on.

Relatedly, the way you have an AI select people and companies to get spot-checked by tax investigators is not to show investigators the AI scores, but mix in AI suspicions among a stream of randomly selected people.

Relatedly, the way you have AI involved in medical diagnoses is not to tell the human doctor results, but suggest additional tests to be made. The “have you ruled out lupus” approach.

And from what I’ve heard the medical profession actually got that right from the very beginning. They know what priming and bias is. Law enforcement? I fear we’ll have to ELI5 them the basics for the next five hundred years.

European police is very much armed. Also the UK has armed units even if your usual beat cop is limited to pepper spray and a baton or whatnot.

Elsewhere police regularly carry pistols, but are also trained in how to not use them. In my state there’s even an assault rifle (actual one) in every police car. Decades pass without anyone getting shot.

I think it’s a blend, in my example the police would bring them into custody, and then trained people work with them after that working out what happened and working with the justice department.

Nope. Police is not trained to deal with e.g. a psychotic person seeing zombies, if they try to take them into custody they’re only going to make things worse. It’s fine if police are first to the scene, but they should be trained enough to a) recognise that the person is psychotic, not actually threatening anyone b) call for backup from the people in white coats with haloperidol shots and c) shoo away bystanders. Perimeter duty. Yes, after 2 1/2 years training you’re on perimeter duty get used to it that’s your job.

The US approach to a paranoid schizophrenic scared shitless seems to be to make it worse by laying siege and throwing flashbangs.

There are many things that police aren’t needed at, like domestic issues, but there are plenty we do need them at too.

That’s probably the bulk of what beat cops are doing over here, short of investigating noise complaints on behest of the municipality and documenting traffic accidents, car thefts, maybe a break-in, whatever. Which is also why they always, and I mean always, come in male/female pairs.

If you can’t see how that’s a (small) puzzle piece of a long-term strategy then I can’t help you either.

GDPR etc. are the tip of the iceberg when it comes to EU regulations. Plenty of them apply to stuff where the EU is world-leading, also very far-reaching ones. Compared to REACH, GDPR is quite low-impact indeed.

Donations are a tiny fraction of Mozilla’s income. Firefox and related projects are their money earners for their actually charitable projects, pulling in at least half a billion or so a year.

Not saying that the CEO pay is adequate or something, but your take is literally ignoring the article you yourself quoted.

They still might very well be over-represented because looking at decompression errors doesn’t isolate the CPU, could also be the disk, or RAM. Or even download though that tends to have independent checksumming. And it might not even be he components it could be cosmic rays, if you run code often enough on enough boxes errors are unavoidable, at least on hardware that isn’t space-grade and/or buried underground.

Mir ist generell kein Wort bekannt, bei dem sich ein a zu ä, o zu ö oder u zu ü ändert, wenn es mit einem anderen Wort kombiniert wird.

Nach Gold graben -> Goldgräber sein.

Wobei hast schon recht der Umlaut kommt nicht durch die Zusammensetzung zustande sondern durch die Nominalisierung von “graben”. Ansonsten kommen Umlaute noch bei der Steigerung von Adjektiven vor (alt, älter), sowie Pluralbildungen (Gans, Gänse) und beim Präsens vieler starker Verben und auch der Konjunktiv ist mit Umlauten durchsetzt und das war’s dann glaub ich auch schon.

There’s plenty of designs out there that you can load onto an FPGA or, funds permitting, send off to a fab to burn into silicon.

Giving you, if you were lucky, VESA graphics and maybe a mouse pointer because XFree86 somehow insisted on being told whether you have a PS/2 or USB mouse. 3d acceleration only with nvidia and that required manual installation because nvidia never provided anything but blobs. IIRC ATI drivers were simply non-existent (didn’t have an ATI card back then), that only changed when AMD bought them. Whippensnappers won’t believe it but once upon the time, nvidia was actually the company to go with when running linux. And Epic didn’t hate Linux yet, UT2004 came with linux binaries on the dvd.

There’s one two guarantees in life: Death and taxes. Real currencies are the ones you can use to pay the latter.