Posts by Elongated Muskrat

'Skeleton Key' attack unlocks the worst of AI, says Microsoft

Re: All this proves is that ...

Perhaps we are all automata, controlled by tiny Gordons, and they just hide out of the way, having a cuppa from Bertrand Russell's Teapot whenever neuroscientists are looking.

Re: All this proves is that ...

Probabilistic models can, of course, be incredible intricate and can entail deep complexity. Like, errm, quantum physics, for example.

The fundamental principles of probabilistic statistical quantum mechanics are surprisingly simple, and entirely deterministic, in that given knowledge of the system, probabilities of any outcome can be calculated. This is where it fundamentally differs from LLMs which work in such a way that the starting conditions and parameters are not known, which makes them essentially non-deterministic.

The compare with QM; take the double-slit experiment as an example. Fire one electron through an aperture made of two closely-spaced slits, so that it hits a phosphor and emits a blink of light; you cannot predict where it will hit, but you can assign a probability to each point, based upon the wave-function of the electron interfering with itself (oo-er matron, oh no she didn't, quiet at the back, etc. etc.).

Fire a billion electrons through the same slits, and you can predict pretty much exactly the pattern of light that will be emitted (which will be a series of interference fringes). When you get to the scale of the physical macroscopic world, where the number of particles in play in a system is many orders of magnitude bigger than this, such as the arrangement of gas molecules in a room, and you can work out the probability of any arrangement (such as all the molecules lining up against one wall), and see statistically what the result will be. The outliers, such as all the air molecules being in the corner of the room will be expected to occur once in many times the lifetime of the universe, but statistically, not technically never.

How does this relate to AI? Well, repeat a query to AI a million times, and you're unlikely to get a useful statistical pattern. Execute that query just once, and the result is pretty much entirely non-predictable. In most real-world use cases, it is helpful for computer software to act predictably.

Re: All this proves is that ...

There's no objective difference between the output of AI and something written by a human making stuff up on the fly.

The difference between well-researched technical prose produced by a human, and the stuff AI churns out is night and day.

Re: Another Arms Race

The idiom that describes this was coined decades ago: GIGO.

Garbage in, garbage out.

Re: Some of us ...

Some of us were lucky enough not to be exposed to harmful books as children.

Great, so next up, expect to see LLMs start a religious war and/or arbitrarily decide that humans with one set of physical characteristics are inferior, and commit genocide?

Re: Another Arms Race

c. Stop pretending that LLMs can answer questions from humans in a meaningful, controlled, and predictable way, and admit that they are very hit-and-miss in their usefulness.

Re: All this proves is that ...

A large number of humans can't do this at all, and I'm pretty sure that nobody can do it reliably, and reproducibly. Bear in mind that the Turing test is based on a though experiment, and is subjective (being based on the opinion of the tester), and thus not an actual useful measure of anything.

Re: All this proves is that ...

Two things:

1) Just because something is non-deterministic, doesn't mean it has agency - this is magical thinking and is basically the basis for animist religions.

2) The problem of defining intelligence, or consciousness lies in the paradox that to fully define something, the definer must not be part of the system being defined. To define consciousness, one would have to define one's own consciousness' ability to define itself, which is a recursive definition. This, however, doesn't mean we can't identify things that aren't conscious, it only means we can't fully define consciousness. A toy that has been fed a large amount of data so that it produces superficially intelligent-looking output under some circumstances is so far from exhibiting consciousness-like properties that it is laughable to suggest that it does, or might. All that you can say is that, due to its nature, it is computationally non-deterministic in its output - you need to know all the training data, including every single previous input, and source of entropy (such as RNG seeds) in order to replicate, and thus predict, behaviour.

Re: Bricks

Except a wall made from bricks is useful, but a wall built by a LLM would probably have courses that don't line up, and too many fingers on it.

Microsoft yanks Windows 11 update after boot loop blunder

Re: Don't blame MS

Here we are, in 2024, and I'm still seeing the same tedious "Apple vs Microsoft" arguments. Guess what? They're both big companies that want your money, and neither is your friend.

Businesses use Microsoft, or, if they're a design house, probably Apple, because that is the tech stack their business processes are built upon. Arguments about "move to Apple" or "move to Linux" or move to X, Y or Z seem to miss the point that this involves change and change management is both costly, and uncertain. Expense and uncertainty are two things that businesses avoid, for some reason. Can't think why?

Airbn-bye: Barcelona bans short-term apartment rentals for tourists

Re: At a minimum

I think another reason is that in much of the western world, the type of governments we have had in the last couple of decades have tended towards the "fewer regulations" type, and not the "more regulations" types. This links directly to the tendency towards a more neoliberal economic model (fewer regulations, more "austerity", smaller state, more free rein for capital to draw money to itself), which favours big operators like AirBnB (and Amazon, Google, Apple, Microsoft, Tesla, et al). This is in large part down to the fact that we measure economic success by the ability of corporations to grow and become richer, and not by the even distribution of wealth throughout society, so we can claim we are more "successful" whilst the vast majority of people are becoming more miserable and impoverished.

We need a willingness for legislators to regulate, but that is often in direct opposition to the people funding them. I'm open to suggestions for how to solve this seemingly intractable problem, short of armed insurrection (which superficially seems like it might be a good idea, right up to the point where you end up as one of the ones up buried in a mass grave).

Re: Alternatively….

I think it's a bit more subtle than this. AirBnBs undercut hotels, principally because they are not strictly regulated in the same way, so have lower costs. This in turn drives down the prices of hotel rooms, as they have to compete fort his business. This makes them less profitable, so they go out of business, the end result being reducing capacity of "proper" hotels, then the AirBnB operators realise that they no longer have any competition from hotels, and can ramp their prices right up, which has a knock-on effect of attracting more people to do this, reducing housing stock, and driving the cost of that up, too.

We need to get serious about treating residential property as homes, and not as an investment, or as a business. People need somewhere to live, and in places where there is a shortage of that, short-term lets should be severely limited (or banned outright in extreme cases), as should owning a second property in the area. This could easily be subject to various allowances such as inherited property, and homes that are unoccupied because someone has moved in with their partner instead - give a limit on how long they can be owned before attracting increasing taxes, for example, giving an incentive to release that unoccupied housing back to the community that needs it.

Re: At a minimum

This is a good point, and I think it's important to draw a distinction between people letting out a room for a bit of extra cash, or people letting out their property when on holiday, and people letting out a property solely as a short-term let.

I think for the former, licensing and a limit on the number of nights they can be let in a year (say 28, for example), along with proper enforcement, would prevent this system from being abused by people doing the latter. For them, they should be licensed and regulated in exactly the same way as a hotel or B&B is. The problem really seems to be that there is no regulation at all, and this almost certainly has come about because those in place to create those regulations are either profiting from this themselves, or are subject to heavy lobbying from those who are. The adage of "follow the money" always holds true.

You could read my other post below, and see that this is not what I am saying at all.

If anything, uncontrolled AirBnB letting drives both rents up, and hotel prices down, hurting both local people, and local businesses. What I am saying is that we need regulation to limit the number of residential properties used for short-term lets, thus removing them from the housing stock, but also allow for building of purpose-built short-term lets, or conversion of empty buildings in areas where there is not a demand for accommodation, such as city-centre office buildings, which are in a great location for a place to stay for a few nights when sightseeing, but are not suitable for people to live in long-term as there is no prospect of peace and quiet, or, generally, access to open spaces.

The thing is, it's not an "either/or" situation, and I think you're deliberately misrepresenting what I am saying. Whilst there is a housing shortage, there should not be a laissez-faire attitude towards unregulated short-term lets, as they are only good for the people letting them out. Tourism isn't going to fall off a cliff if people have to pay a few more quid a night to stay in a hotel instead.

There are plenty of legitimate reasons for people to own two properties, as residential properties for themselves or family, there are fewer reasons to own three, even fewer to own four, so that by the time that you get to someone owning an entire block full of properties, and letting them out nightly at hotel rates, there are no legitimate reasons at all, unless you happen to count profiteering from driving down the supply of housing stock as legitimate.

There are two extremes; owning (or having access to) no property whatsoever, and being beholden to the whims of landlords, or owning all the property, and controlling the housing of everyone else. Obviously the "happy point" lies somewhere in the middle, where nobody is homeless, and nobody is price-gouging the rental market.

Exactly this. I've long thought that the "short term let" sector needs some heavy regulation.

It's literally a cash-tap for anyone who can afford to buy up a property, do it up for rental, and let it out on AirBnB. Rather than a residential let at, say, £1,000 PCM, they are renting the same property at often over £100 per night to tourists. The effect of this is to not only to push local rents up, towards that £3,000 PCM, but to push prices down for hotels, thus driving them out of business.

However, the flip-side of this is that tourism brings money into places, and tourist accommodation isn't a bad thing per se, so what is required is proper regulation - things like a limit on the number of residential properties that can be converted to short-term lets, licensing systems, to control numbers and distribution, purpose-built short-term lets that don't eat up the housing stock (convert empty city-centre office buildings, for example, where it's not a good place for people to live permanently). Introduce limits on how many nights a year someone can rent out their own property, or spare room, so that "normal" people can still rent their place on AirBnB when they're on holiday, but people can't buy properties and let them permanently as a source of income. Back these measures up with real, and heavy penalties, for those who try to work around the system, and not just a slap-on-the-wrist that can be absorbed as a cost of doing business.

If you're using Polyfill.io code on your site – like 100,000+ are – remove it immediately

Re: Third party scripts

Exactly. Add to that anything that wants to run a script to show you an advert, because that's pretty much the number 1 malware vector.

I'd say that probably 99% of adverts out there cross at least one of these red lines, and that's why I block them all, both in the browser, and at the network level at home by blocking advertising domains.

Re: That's why I hate java script

I don't buy the "site improvement" stuff. There are plenty of successful design idioms out there, and you don't need to track your users to copy them.

Conversely, how often do you see a site do a complete layout redesign? How many sites spewing google analytics at you do this, ever?

I think the answers to those questions would enlighten you to how much of a bullshit factor applies to this argument.

Re: Tip of the Day

Yeah, I've corrected the browsers on my work machines, will do so on my home machines when I finish work, and also add to the Pi-Hole.

Re: Third party scripts

I'd say the same, but with the provision that I'm not fine with ads that are intrusive. If I don't want to look at your ad, it's my choice, and forcing me to do so doesn't endear me to the product you are trying to flog to me.

I'm looking at you, unskippable 30 second video ads for the same thing EVERY SINGLE TIME, complete with loud sound, and tiny, teeny, microscopic close buttons on mobile apps.

Re: That's why I hate java script

If they do that, then they are taking (legal) responsibility for the content of that script. On their heads be it.

Re: That's why I hate java script

Why does my browser want to run scripts from googletagmanager.com when the browser bar clearly shows that I'm visiting theregister.com?

That's easy - I have NoScript installed, and that domain is forbidden to serve up scripts to my browser, so it doesn't. I've yet to encounter a site that is broken as a result.

Since joining NATO, Sweden claims Russia has been borking Nordic satellites

Russian history is not so much littered with public hangings, as people carelessly falling out of windows, meeting untimely ends in basements, and accidentally eating something they shouldn't have. I think cousin Vlad is probably very wary by now of only ever being on the ground floor of a building, and having someone taste all his food for him.

BOFH: Why's the network so slow?

Re: Mmmmm....

Sounds to me like that your audit table was a heap when it should have had a clustered index on audit ID, leading to table scans on key lookups, but what do I know?

Your bottleneck there is most likely to be on disk access on your database server, not network, and proper fault analysis would have shown that.

Consultancy fees are payable in beer.

Mozilla defies Kremlin, restores banned Firefox add-ons in Russia

Re: Well...

The FSB / GRB / Kremlin-funded TLA of choice pay him per click.

BOFH: An 'AI PC' for an Acutely Ignorant user

Re: Virtual coloured penclis?

Apologies in advance for the full pedant mode, but "isopropanol alcohol" is a tautology; it's "isopropyl alcohol", or just "isopropanol"; isopropanol being an isopropyl group attached to a hydroxyl (-OH) group.

In the same way, you'd not say "ethanol alcohol".

It's many years since I worked as a chemist, but the little things like this still make me twitch...

Re: Blinkenlights

I still remember the time when barcode scanners started becoming more prevalent (and didn't cost hundreds of pounds any more). I took one that was in the machine room of the place I was working at the time and randomly scanned a barcode that was on the side of a nearby cardboard box, that had previously held reams of paper or similar, to test it.

The scanner beeped, and the content of the barcode appeared in the Notepad window:

*BARCODE*

Kudos to whoever had put that on their boxes, I suspect that somewhere there was a manager who had insisted all products were barcoded...

Re: Virtual coloured penclis?

I've never seen one of those that didn't have a smeared squiggle in one corner where someone accidentally wrote on it in permanent marker and then tried to wipe it off.

Re: Virtual coloured penclis?

Well, I, for one, won't be driving a Turbot, unless they're fully amphibious.

Japan's space agency helps to target advertising with satellite photos of crops

If it brings down the price of okinomiyaki

It's all good.

Norway's sovereign wealth fund aims to zap Musk's monster Tesla pay deal

Re: Take my cars and go home

All he needs is "one hundred beeeelion dollars". Isn't that right, Mr Bigglesworth?

Re: Profits Aren't Solely Due to Musk

To be fair, at this point, the only people who deserve a massive pay-out are those with the balls to short-sell Tesla stock.

Re: Norway better prepare itself.....

I'm willing to bet that Teslas don't operate well in Norwegian climactic conditions, either. Temperatures in winter easily drop to -10°C in many places there. EVs without noticeable panel gaps (allegedly) are available as well.

I didn't touch a thing – just some cables and a monitor – and my computer broke

Re: Don't forget

The correct answer to this, is "yes", before turning around and walking away.

Re: Yeah, sure. Nothing changed. Pinky promise.

Or, it was someone from her company, who knew her, also knew she was clearly a salesperson (i.e. habitual liar), had seen her spill the coffee in it, and, quite rightfully, when faced with the evidence (a laptop full of coffee), told her that she was lying, and refused to take responsibility for her lie, when she claimed it "just stopped working"?

Meta faces multiple complaints in Europe over plans to train AI on user data

Re: The AI creators are free-loading on other peoples work.

You claim in the same post, that what you are describing is just gzipping a book with lossless compression and also LLM parameter sets are a few GB, out of the tens of petabytes they are trained on. This doesn't actually demonstrate whether this is lossless compression, or not. Tens of petabytes of written material might well compress down to that size, in a lossless fashion, given the right algorithm.

In any case, whether it is lossless or not is a bit of a red herring. I could reproduce a grainy and lossy JPEG version of the "golden arches" and stick it on the front of a burger joint, and still expect to get sued out of existence by McDonalds. I think the actual test in law (and of course, IANAL) would be more around whether a reproduction retains the essence of the original.

Re: "Rather than Meta asking for consent, users must opt out of the default data slurp"

Any organisation can declare pretty much any processing to be in its (or a 3rd party's) legitimate interest, unless it involves processing of the protected categories of data

Whilst this is true, I'm of the opinion that unless an organisation has to process my data, then merely wanting to for profit, isn't really legitimate. My own interests in having them not do this overrides their interest in making money off my back, and they should therefore have an explicit opt-in, if they want to do this.

Re: "Rather than Meta asking for consent, users must opt out of the default data slurp"

What is Meta's legitimate interest in my personal data? I have no relationship with them as a legitimate processor of my data, and no agreement (beyond the T&Cs of Facebook) with them to do anything at all with my data. Just because they had my data before GDPR existed does not mean they are not bound by it.

AFAIK "legitimate interest" means usage by organisations that have a statutory requirement to hold that data, such as public service organisations, of which Meta definitely isn't one.

In other words, for "legitimate interest" to apply, Meta must pass the "three part test":

Purpose test – is there a legitimate interest behind the processing?

Necessity test – is the processing necessary for that purpose?

Balancing test – is the legitimate interest overridden by the individual’s interests, rights or freedoms?

I think it's pretty clear that any "legitimate interest" claim fails on the balancing test, in that it is clearly not in my interests for my personal information to be freely collected, analysed, and sold, for profit, without any oversight from myself.

Opt Out Now!

A spokesperson for Meta told The Register: "We are confident that our approach complies with privacy laws, and our approach is consistent with how other tech companies are developing and improving their AI experiences in Europe."

For starters, requiring you to opt-out of your data being processed is a GDPR no-no. I managed to do so, and they didn't make it easy - click on a non-obvious link in the middle of the "changes to terms" notification they send you (mixed in with your other notifications), and then fill in a form (including submitting and verifying your email address) giving them your grounds for objecting. I told them it was on the grounds of violation of my privacy, and they responded that I had been opted out "successfully", but I should not have had to do this. It seems like a clear GDPR violation to me.

More layoffs at Microsoft: What's really going on here?

Re: Mixed Reality group

It's a classic example of a solution looking for a problem.

Computers are good at solving well defined and delimited problems quickly.

Humans are good at processing and understanding natural language, innovation, and creativity.

LLMs are somewhere in the middle, where they can do neither, but exist in the very small overlap between things that don't need to be done properly, and things that don't need to be understood properly.

Re: Stab in the back

I'd say that £55-£75K range was a senior dev's salary, and a tech director (on an actual board) would be on twice that, but what would I know?

Re: Mixed Reality group

How's that 3D TV holding up? Anyone? Anyone? Bueller? Anyone?

Checkmate? AI's pawn-pushing prowess proves partly pitiful, partly promising

Re: So they can't play chess without waffling.

It should come as absolutely no surprise to anyone that a LLM can learn chess about as effectively as a toaster, or a piece of cheese. There's a lot of magical thinking going on in the field, and that starts with calling statistical models "AI" in the first place.

Twitter 'supersharers' of fake news tend to be older Republican women

Re: Cesspit

You don't actually have a defence for the far left blah blah blah

You literally don't know what a straw man argument is, do you?

Your claims have absolutely nothing to do with my post, and I see no point in following you down your rabbit hole. It's enough to point out that you made a straw man argument, and that it was a straw man argument, and also demonstrably a false one. You said you couldn't think of an example of something happening, and I gave you a very strongly historically documented example of it. Beyond that, I couldn't give a monkey's ass about your claims.

My point is that this sort of research brings out exactly the sort of rubbish comment you have posted, because people will do and say anything before accepting that the things they are saying and doing are provably wrong. You can argue yourself round in circles all day long, all you are demonstrating is that you can argue yourself in circles all day long. My observing you doing this is in no way an indication that I wish to join in your circle-jerk.

Please give me a list of all the US presidents who were not practising members of the Christian faith. Go on, I'll wait.

The fact is, that in the US, if you don't profess to possess religious faith of some flavour, it essentially makes you unelectable. The opposite holds true in the UK, where more than half of the population registered themselves as having "no religion" in the last census. It's one of the very few ways in which the UK is still a progressive society with all the backwards steps we have taken in the last decade-and-a-half.

Re: True statement

Mainstream media has been telling people what to think for ages now.

All media does this, it's just that mainstream media has at least some semblance of regulation that intends to prevent it from being pure disinformation, unlike the alternative.

It's therefore legitimate to wonder whether anyone espousing the alternative has an agenda to push such disinformation, in an unregulated and uncontrolled manner, for their own nefarious purposes.

If you don't trust the "mainstream media", the natural response would be to push for greater transparency, and greater regulation of the media, in order to promote truth over disinformation; oddly enough, these are exactly the things that those who bleat on against "the MSM" espouse, thus exposing their own disingenuity through their own words.

You might as well wear a T-shirt with [CITATION REQUIRED] written on it.

Re: Did you see Ken Burns' commencement speech to Brandeis graduates?

Social Media has effectively turned many people into a "me me me " rather than "we we we" group of sheep.

I think you have just neatly described "right vs left", aka "the individual" vs "the group". The problem is that anyone advocating for one side over the other tends to characterise the other as an extreme. Both extreme individualism and extreme communism are harmful. To those on the right, everyone on the left is Joe Stalin, and to those on the left, everyone on the right is Adolf Hitler. It should be apparent to anyone with a brain that both extreme ideologies have obvious flaws.

I've met plenty of people who are quite hard to either the left or the right, and I have to say, on balance, those on the far-left tend to be less obnoxious (just) than those on the far right, but I wouldn't want to be either.

Most people actually fall somewhere in the middle. The problem is that, at the moment, in the UK at least, whilst the majority of the population falls somewhere slightly to the left of centre, our government is somewhere quite a long way off to the right, playing the same "demonise the minority" games that history has shown us do not end well for anyone. We are seeing far right-wing politics dominated by lies because the extreme element of the right wing is in power; however, if we had an equally far-left government, we would likely be seeing a preponderance of bullshit from that direction. The voice of sanity is to call for politics to re-centre itself to some sort of social democratic norm. One day, we might even get some electoral reform so that the system can't be so easily exploited by those willing to tell lies to get what they want.

Re: It is more than that

This is actually a very good idea. The problem with things being shared and re-shared is that they can grow exponentially, and those things that reaffirm one's beliefs tend to get shared without thought (that's human nature, and not a left vs right issue). What happens is a positive feedback, where things get reposted faster than they fall out of view, whereas, this needs a negative feedback to balance this.

The problem with social media, is that the positive feedback is amplified by those things that can be monetised (paid for posts getting shown more, for instance), and it is actually against the interests of the social media platforms to throttle misinformation, because they stand to gain from any monetisation.

Things that people engage with more, completely regardless of the truth of their content, lead to more advertising impressions. It is not in the financial interests of the platform providers to implement measures to de-platform dangerous misinformation if it is popular misinformation, and they actively oppose any measures to legislate for this responsibility. Money talks.

This isn't even necessarily a deliberate policy, but is a natural emergent property of the idea of social media as a "neutral carrier". This is why legislation is required to regulate social media content, but then there is the careful line to tread between countering dangerous untrue but convincing misinformation and creating dangerous censorship of true, but unpopular information. What is needed is a mechanism to remove clearly untrue material from general circulation, whilst allowing an open process where the content that has been removed (where it isn't downright illegal) can be reviewed by the general public, to prevent abuse of such a system.

After all, everyone is (or at least should be) happy that there is a need to prevent and quickly remove things like child pornography, but even when you move onto things like "terrorism" you run into the "one person's terrorist is another person's freedom fighter" issue.

The underlying problem is that the internet is still very much an emerging platform, and things that would be deemed to be completely socially unacceptable, or illegal, in meatspace are utterly unregulated in cyberspace, even after 30+ years of the web as we know it. The fact that the platform is trans-national has a lot to do with this, but regulation in larger jurisdictions, or in the jurisdictions in which the main social media sites are hosted would go a long way to ameliorating this.

Re: Cesspit

...or, it's nothing at all about "right vs left", but actually about what is observable to be objectively true; in this case, that older right-wing women in certain US states are, for some reason, much more likely to spread things that are objectively false on Twitter?

This is exactly my point; anyone falling even close to that demographic is here tying themselves in knots trying to spin this as "yes but the left". Just like your comment above. I suspect you even know this, as you chose to post your comment anonymously.

I've yet to see a right leaning group walking through a residential area at 4am banging drums and shining lasers and bright lights into windows to wake the occupants or painting graffiti and smashing windows of local businesses and university buildings or setting fire to apartment blocks.

In the 1930s, gangs of German right-wing supporters went round smashing windows, waking the occupants, daubing graffiti et al. I expect you're about to claim that Nazis were socialists now, or perhaps you have't studied enough history to have heard of Kristallnacht?

I know I'm inviting a cry of "Godwin", but anyone about to do so should consider that actual valid comparisons to the Nazis are exempt from reductio ad Hilterum, and this particular AC is inviting that comparison by claiming to never have had it occur to him.

Note as well, that I'm not supporting any such actions being taken, whether they are by people on the left or right of the political spectrum, I'm just pointing out the sheer nonsense and bone-headed stupidity of this particular straw man argument.