Re: It's not just phones
> USB sticks can normally be prised open and the flash chips cut up with pliers
Not going to be much left to mine for precious resources after you've done all that.
5065 publicly visible posts • joined 9 Nov 2021
> I'm not letting anyone have them until I know all of my data has been permanently removed.
You aren't going to have a choice, eventually someone will get them off you (even if that is literally prising them from your cold, dead, hands - or just during the house clearance).
Perhaps you need a better approach than hoping some random commentard will come up with a sensible suggestion? And act upon it whilst you still have the faculties to do so.
Targeted advertising - the only way to stay safe, to keep your privacy, is to be exactly the same as everyone else. Think the same, shop the same, browse the same and you can sink into invisibility, just one of the indistinguishable masses. Perfect fodder for the mercantile cannon of industries fighting for their rightful portion of your life: Welcome, my son, welcome to The Machine.
Or fire up your PiHole, cast Chrome to the wind, cry "For Harry" and delete cookies on exit. They may take our cycles with their outrageous scripts but they shall never have our Topics.
"Your model WILL do something unpredictable, and we can't predict when"
Not some wishy-washy line like:
> it's not outside the realm of possibility that they could act in ways that are difficult to predict.
As the ML these days is pretty much all based on neural nets[1] (including GANs) you'll end up with a magic box that is pretty much guaranteed to do something unpredictable[2]: why do you think we have so many cautionary tales about Djinn?
[1] is *anyone* doing anything different, like running any form of rule inference learning and generating explainable systems? Not that the created rules are guaranteed sane, but you can at least read them out during the court trial.
[2] are you absolutely sure your training and data conditioning will ensure the model will never see an input out of its expected range? It's not like feeding numbers into a simple equation, a model can go totally Bursar on you. Look up "glitch tokens", especially if you are offering a cheap and quick - sorry, "affordable" - model-building service ("save money by not starting totally from scratch this time").
Kodak PhotoCD, launched just before Win 3.1 IIRC
Take your photos/prints down to the local copyshop/film developer - or be really sure that you wanted to share all the shots on a new roll of film and tick the box on the yellow envelope that came with the film. Postal orders or cheques enclosed for payment or write your account number in the space provided.
The whitepaper includes some interesting statements, examples include (annoyingly, can't copy'n'paste direct quotes from the PDF on this device to give the full effect):
Pointing out that there will have to be a way to verify the logic in the code of a "smart contract"; this can not possibly make the system expensive to use, as we all know that automated code proofs are easily available and, if not, human programmers can do code checks for barely over the minimum wage. (/s - superfluous, but may help if someone really is holding up the "will review code for food" cardboard sign).
They point out that credit cards exist, but apparently they will only protect the supplier of the goods, not the consumer. Um, I know making a chargeback can be a pain, but a CC giving *no* protections to the consumer when used for distance shopping? Clearly, the banks authoring the whitepaper know more than I, but... yikes?
PS
Haven't spotted any mention or reference to *how* the "smart" bit will be coded (if you have, please correct me) so looks like Javascript is still on the cards (well, it does make eating JSON data easier, what could possibly go wrong with an eval(readfile())?)
I read it as closer to putting a hold on the money whilst it is still at your bank (so no extra trusted third party) and transferred in payment when the conditions are met (delivery of goods); one also hopes that the conditions are also set to allow cancellation of the order and/or a maximum time for completion, after which the hold is released an you get control over the funds back.
This way, your bank gets all the transaction fees (including when the supplier calls to verify the hold is in place) nothing for that greedy escrow agent.
UNFORTUNATELY although the above fits the description given in the article and could actually be a useful addition to banking services, the actual whitepaper starts by prattling on about digital currencies and stablecoin, so it isn't going to be anything sensible and will probably end up with overly-complicated smart-contracts that are coded in Javascript and can be gamed in all sorts of fun ways. Hmm, if I can cause *this* "contract" to trigger now, which will trigger Fred's contract, but mine is so badly code (oops) that it garbage collects, meanwhile Fred's has completed and my other contract has swept in, got the funds, raised some interest, released it back to my first contract which is now ready to continue...
Also used work on Windows Embedded boxes that boot off CF cards (that kit is still in service and should have years left in its life - no, it isn't connected to the Internet).
At home, still have a perfectly functional CF bootable mini-ITX mainboard system, ready to be put back into service.
The problem nowadays is that IDE is only one of the modes that CF supports and the later cards stopped supplying it. So have gone from "any of the CF cards work to boot from" in the days of 512MB cards to "you suddenly pay a massive surcharge for a card that actually follows the whole spec" which made, say, a 4GB CF too costly to be worth it.
And three guesses whether an up to date copy of the OS that used to fit in 512MB will now even fit onto the 4GB card if I'd bought it! Annoying, as the 512MB card is in perfect nick (it only got read at boot, never written to during normal use). Ah well, at least SSDs are more affordable now than bootable CF cards.
Huh? Why do you want to remove that one?
> "display file icon on thumbnails,"
Ah, now it makes sense.
Remove those two and you'll never see icons for anything that can be thumbnailed so you won't spot the moment that Microsoft overrides your choice of image viewer. Again.
"IrfanView? Bleugh! You want a proper Microsoft Media Experience!"
> Right now, the only workarounds for stopping the high CPU utilization problem is either restarting the device or for the affected user to sign out. Locking Windows won't do the trick, Redmond wrote. The users must sign out.
So using Task Manager to kill File Manager has stopped working under Windows 11 as well?
And why would anyone expect locking Windows would help? Doesn't anybody else lock Windows when they go for a cuppa, fully expecting any tasks to continue?
(Disclaimer: zero practical experience using Win11; for all I know, TM has been neutered and locking now means deep sleep!)
> the alternative would be designing a milspec controller that ends up costing >$1m a unit and does the same thing
The story is that they *did* have a "milspec" controller supplied by the contractor, at an obscene cost, until a savvy submariner pointed out to the boss that it was actually just a video game controller (probably rebranded, that is what cost so much). Won't swear that they actually now use an XBox device.
The controller is only used for pointing the modern replacement for the periscope (and no doubt is not the *only* way of doing so - they may even still have those funky folding handles and can wear their caps backwards as a backup, before resorting to "ratings and binoculars").
(Tried to find my source to post the URL but search results are currently flooded with thr Titan story).
> He is able to beam himself at will
EM activates the teleportation by getting into the driver's seat, closing his eyes and meditating for five minutes: when he opens his eyes, ta-da they have arrived!
Meanwhile:
As soon as EM drifts off to sleep ("meditating") the *real* driver takes over (amazing what you can do with a Tesla paired to a bluetooth steering wheel from an old Wii game). When they arrive, all Tesla and SpaceX clocks are wound back to five minutes after they left (if it was only a two minute journey, they get wound forwards) and EM is gently prodded awake again.
This is how EM manages to do his amazingly long work days *and* explains why he is so happy to make those claims about "FSD is only 3 months away" or "you can buy a Semi with delivery in a year": he is getting lots of sleep and, according to his watch, it is currently an unseasonably warm February in 2017.
Don't worry about the employees: the worst they have to worry about is ignoring the factory wall clocks and being sure to have the company app running during meetings with the boss, showing EM time. Okay, there are a few odd days when they end up eating three breakfasts in one day full of meetings and they are super-careful around launch days (when they balance things out by having lots of two or three minute drives around the launch site for each fifteen minute drive from hotel to site).
Are there parts of RHEL that are *not* covered by the GPL and without which a fully RHEL compatible version is not possible?
They need not be major parts, in the grand scheme of things, but if there any little bits which may really be restricted by RH's actions that would be enough to prevent any rebuilds from being 100% bug-compatible with RHEL. A point which RH would be sure to emphasise.
If such a situation exists, *we* may all know that those components are irrelevant to 99.999% of Users, but even a few differences would be enough for RH sales & marketing to leverage into full-blown FUD ("you are unique, you are one of the 0.001%" would probably work on most CEOs).
All they have to do is set the perceptions about remixes they don't like, not the reality.
I'm happy with using the word "hallucination".
Especially in the sentiment that: "It’s not that they sometimes 'hallucinate'. They hallucinate 100% of the time, it’s just that the training results in the hallucinations having a high chance of being accidentally correct."
If you don't like "hallucinate" about the only other word I know that comes close is "stoned". Groovy.
> Then we'll probably get an AI winter
Oh joy, *another* AI Winter?
WTF can't we just get on with working on these problems without some - people - suddenly deciding that they can hype[1] the hell out of the thing and then embarrass everyone when they fall on their arse!
[1] Yes, I know that science funding proposals end up having some hype in them, but I'm referring to ChatGPT levels of hype: the general public doesn't hear about funding proposals on the TV news.
> LLMs are a new thing.
True in that their actual existence is new - but that youthfulness is down to economics alone (i.e. paying for the cycles required).
But how to make them has been known and thought about for decades.
> nor our education, has prepared us for
Ah, THAT is the nub of the matter. These things haven't been widely taught - and now that they "have entered the public perception"[1] 95%[2] of the available materials are utterly useless to Joe Bloggs The Lawyer; if only he that.
[1] i.e. as usual, everyone who was talking about them was ignored: science outreach is a Very Good Thing but falters because it is, of necessity(?), voluntary on the recipient's end
[2] wild optimism?
[3] I can point you at a lot of good academic or technical material ("whooosh" says Joe) or lots and lots of total bollocks and frankly dangerous "do this, it will redefine your business" twaddle on YouTube.
When people do bloody stupid things like this in other fields, the next few years are spent shouting at and suing each other in court and the waters get muddied, things get settled out of court and/or restrictions are placed on the details of what actually went wrong.
Here the situation has gone straight to a judge, all of the details are in the record for us to see and the judgement is startlingly comprehensible; he almost came right out and said it as simply as "Use the proper tool for the job and that choice is your personal responsibility".
Now, hopefully, the company lawyers will be looking at what their company proposes and be saying "You are using an LLM here? Nope, not going to sign this off, not risking myself in court."
The fines are not going to financially scupper the law firm, but their imposition has made it absolutely clear that the behaviour won't be ignored, by this or other courts (in the US): they were let off light for the first offence, but everyone has been warned.
> It's bad programming, no more, and no less.
The programming is fine (the code works and performs the maths required of it)
> model being used is fundamentally flawed
The model is fine: it is built upon the maths started back in the 1950s and does what couldn't be done back then: use large numbers of cycles to train on large amounts of input. It builds a correlator.
What *is* wrong is *applying* any of the above for a any purpose other than amusement.
Treat just like the Radium Snakeoil Sales - radium itself is not to blame, it has its place and can even be useful to humans in its place. BUT that place is NOT toothpaste!
> coupled to ... huge databases that are demonstrably full of incorrect, incomplete and incompatible data
> Please don't glorify it.
Please don't glorify it by even vaguely implying it is coupled to a database - the LLMs aren't using anything more than a honking great pile (not even a carefully organised and categorised by humans who know what they are doing pile) of correlations of "this has been seen to follow that".
Human idiots have piped the output of LLMs into other tools, such as database queries, such as using them for web searches, but describing that as "coupling" the LLM to a database is as meaningful as (apologies for incorrect options usage)
cat text | grep isbn | awk -i reformat_isbn | sqlite all_my_books.db
and describing "grep" or "awk" as being coupled to the database.
A "black box" is something that is magically doing its job - nobody knows how (and sometimes why) it works, it just does.
And we are quite happy leaving it that way.[1]
[1] If you open it, it will immediately stop working and all you'll find in there is a pile of dried leaves ((C) Colin Greenland)
As with any other lists of verboten words we'll start going down the euphemism spiral soon enough (as you point out, they've started out partway down the slide already).
A set of common replacements will be chosen, then they'll become obvious euphemisms and become as disliked as their predecessor, so a new replacement is chosen...
> The phrase "master copy" is yet another mis-application of the word "master".
That isn't the wrong use of "master", it is a "wrong use" of "copy", if anything!
The phrase is simply a shortening of (along the lines of) "the master made this object from which all copies are to be made". The master piece (which may well be a masterpiece, although the copy itself may well be the masterpiece, if that is the requirement to gain one's guild masterhood) is the thing to be copied, whilst the wording "master copy" taken as-written implies that it is the *result* of the copying.
Although the master copy can, of course, be a copy itself: you can create a piece, use it as a master to make copies and then send each of those off to other sites where they will play the role of being the master copy against which those sites will verify their own copies.
You can also have some wordplay, as in "the master copy must copied slavishly", if one is allowed to.
"Master copy" is simply a contraction.
> Or a neurodiverse person (whatever that means) saying how much (s)he resents "sanity test"?
I have a piece of paper that says I am "neurodiverse"[1] - and my go-to code toolkit proudly runs a SanityCheck() every single time. And does not hesitate to print out a message and halt[2] when it fails.
[1] actually, it tells me what my diagnosis is, not just some weird neologism
[2] I first wrote "kills itself" but one has to be sensitive these days
God does indeed have your full details, written down in the big book that St Peter can be seen reading from in all those cartoons; you know the ones, Peter asks "Name?" then looks to see if it is time to use the trapdoor lever or not.
The trouble is, God is really terrible with names, although He can remember a face[1]. So it is important to go to confessional, then He can look down, get a really good gander at you and put a face to all those peccadilloes He'd been hearing about[2]. Oh, and if you can make your sins unique, or at least interesting, that'll help a lot.
[1] And don't suggest adding ID engravings to the book, He'll just sulk. Something about never having got the hang of writing because it came after His Big Six Day Bash and the book is just so fiddly: He only managed to get four words to fit on an entire wall!
[2] Angels just love to gossip about goings on in Midgard[3], if only to give their four faces something to do
[3] Ah, yes, there was something else. He has got Himself some ravens and now self-identifies as Wednesday, so "now everything can be done mid-week and he finally gets a peaceful weekend". Raphael says there is nothing to worry about, but Michael is looking a bit strained these days.
How come the malware people manage to get decent programmers who know how to use meaningful names when some of us start to feel grateful if we get a CalcVal()?
I bet they even use naming conventions so you can at a glance if a line of code is using static, global, local or member variables!
Indeed. As I said recently, multiple time I've signed up for Prime deliberately and used the 30 days, cancelling without any great difficulty; and I plan on doing so later this year, when they make available all the episodes I want to binge watch.
Ok, cancelling means going into the "My account" which is harder than joining - returning a product is about as difficult (i.e. not very, if you read what is actually on the screen), just as buying the product was easier than returning it. Not really a surprise.
Ok, I have hit the wrong button (the big one rather than the link beside it that plainly days "no, I don't want the magical experience" (or something like that) but it has always been possible to catch that before completing checkout.
Amazon certainly promote the heck out of Prime, with big happy buttons and banner ads, but then so does everyone who offers a subscription. But if you read what is presented to you, including what the checkout says (has the p&p just vanished?) then you can escape unscathed. If not, you have 30 days to cancel.
Weirdly, SWMBO has come to me, worried that she had signed up to Prime, and I've had to point out that, no, she hadn't!
BTW this is for the UK website - Amazon apps may be totally different.
All of the cases of "evil AI" given are nothing to do with AI but are simply applying the wrong tool to the problem, mostly[1] down to attaching the LLM to some i/o that it can run rampant with[2].
If I set up an automated tax portal and implemented it by code that just filled your form with Lorem Ipsum and and the Profanisaurus you would rightly say that my service is not fit for purpose and, if I charged you for and posted the forms you could sue. If I use ChatGPT then, again, you can sue. But it is clearly my service that is at fault: neither ChatGPT nor Profanisaurus are correct choices but they are not at fault.
Similarly, I can create any of the other "evil" scenarios without ChatGPT (and probably do it cheaper).
So, legislate against creating *any* stupidly dangerous application; the "AI" part is just theatre for vote grabbing and fear mongering.
BUT OpenAI will never support that, because people will realise that their product is not fit for any purpose and stop using it (after they've been sued enough times, hopefully brfore too much damage is done).
[1] exceptions being the "giving bad advice" situations, such as on mental health; but it is still the wrong tool for the job (or the problem was putting console i,/o on ChatGPT amd letting anyone loose on it).
[2] eg the (apocryphal?) example of letting LLM actually create Facebook accounts
They are just saying *they* wanted a windfall and didn't get it.
There is nothing there to support any idea that Google got a windfall and failed to pass it on: Google just did what everyone expected Google to do.
Google have pretty much had a monopoly on online advertising for a long time now, there are no surprises there for anyone who did their due diligence.