Posts by that one in the corner 5493 publicly visible posts • joined 9 Nov 2021
We know how this one goes: the UK freely shares its IP, the US promises to send back anything it develops from it - and forgets to include "freely" in its half of the agreement.
We remember what happened with the jet engine. And the atomic bomb.
One more horror of Cyber-conversion, when you feel your battery starting to fail: "Pain!" | World Enough And Time | Doctor Who
I understand that Belize is transitioning away from Imperial to Metric units.
Which is a shame, at least from the p.o.v. of some of my code, where you are expected to select Belize to get Imperial and mm/dd/yyyy* used in messages :-)
* Yes, I know it isn't supposed to be, officially, but it isn't totally unknown - that is good enough for this purpose; certainly haven't had any complaints from Belize about this; from colleagues in the UK, yes, after they've exhausted just rolling their eyes at me...
Both cosmic rays and boring old slow alpha particles were found to cause problems, but it was the latter that was occuring far more often (and far, far more demonstrably), for certain devices.
The basic issue was very simple: some of the packaging materials used to make the ICs contained isotopes that released alpha particles; as that packaging was right up against the chips themselves those particles wandered onto the silicon and messed things up (oh, if only there was a ten centimetre air gap between the packaging and the chip; but I suppose that would rather defeat the object of making small ICs, so ignore that idea[1]).
There is nothing at all surprising at finding alpha emitters - they are all around us, every day. Say hello to your friendly neighbourhood uranium (and its pals): there is certainly no shortage of that in the environment (especially if you like to collect antique green glass). If you make up a general purpose packaging material, say plastic pellets[2], in a normal plastic-making factory, you may be picking up rock dust as a contaminant (along with cigaretter smoke - which is also radioactive, btw, general dust and shed human skin cells - icky). Nobody minds, because it isn't grossly affecting the products made with these plastics: it is the 1970s, everything is still big and bulky, the odd rough spot is fine! It may not even *be* from external contamination at the end of the line, your initial feedstock chemicals may be glowing (very) gently in the dark. Basically, unless you spend the money to clean things up, radionuclides are everywhere, man; they are crawling over your skin right now[3]!
Ah money - so unless and until you know that you've got a problem coming from the contamination, you just don't bother to do any better.
So the IC manufacturers (ok, let's be honest - Intel, it was mainly Intel we heard about) had dirty, dirty devices. Then everyone started pointing and laughing, Intel spent the money and cleaned up their packaging materials.
[1] Not that that air gap would have helped; yes, the primary alpha products would be stopped, but the horrid little buggers are things like radon. This short lived (3 and a bit day half life) product of uranium decay is a Noble Gas, so, whistling nonchalantly, it'd cross that gap, nestle up to the silicon, bide its time and *poof* suddenly decay, releasing alpha particles and taking your data with it. Just in case the radon didn't work, *its* decay products have even shorter lives, releasing more alpha (and a few beta) particles before peacefully resting as a thin, thin layer of lead (Pb) on your chip. Isn't Nature wonderful?
[2] Not claiming this is exactly what happened, I don't know the chemisty that was (is?) used for packaging the ICs in question; it could have been pelletised thermoplastic, or a two-part epoxy or... ; nor the specifics of how the contaminants got in; but the general point still holds: dirty be bad.
[3] And doing less damage than all the microscopic bacteria
Ok, I'll bite - who, aside from you, uses/used the phrase "alpha particle degeneration" (your comment was a Googlewhack - well, Duckduckwhack - for the phrase)?
And, aside from a bit of hyperbole (using "degeneration" implies something permanent or long-lasting, so one would think it referred to the physical structure of the memory degenerating, not just the stored charge, as we expect the charge to be a fleeting thing that'll be deliberately changing at a pace) what is inaccurate or misleading about that phrase to warrant using "so called"? Especially when you link to an article that actually describes how alpha particles from packaging cause data errors.
> marketing soon overcame the will of the greybeards
Marketing? Not Finance, looking at the BOM costs? Did Marketing really care about making sure the User knew you were using cheaper DRAM?
Yours, etc, Mr Grumpy-hasn't-his-cofee-yet
> LLM training has next to zero semantic analysis
True. There is no (interesting) semantic analysis explicitly performed by any part of the training, as far as we have been shown.
> hence their method would of involved the creation of a suitable semantic analysis network.
You might hope so, but - got a citation?
The apparent belief about LLMs, certainly what is peddled to us* is that they "magically" (as in, if they are, nobody knows how) create their own network which embodies semantic meaning, which is then spat out at us. Now, if these fine researchers are not eating their own dog food then we would really hope to have that explicitly stated, with reasons why.
Hence, instead of a vaguely hopeful "they would of" (sic) we really need solid citations.
Well, duh.
They just fed all your inputs into the LLM training maw. As they do anything they can get their hands on.
The only "method" they might have needed to "devise" was to take a pre-trained base and feed it all your chat logs as "specialised data" - the same thing the LLM companies try to flog you to create chatbots "personalised to your business practices".
At least this time they are being honest about whether a human ever gets involved to vet the inputs for, you know, accuracy/usefulness/sanity.
> consisting of a 24MHz Cortex M0+ chip, 24KB of flash storage and just 3KB of static RAM, but it was still workable.
Luxury!
There is loads of stuff you can do with that - just look at everything that could be done on, say, an Acorn Atom, an interactive programming environment (admittedly, I blinged mine up with the full 12KB of RAM in order to get colour graphics or a decent game of Space Invaders) but it was still just 1MHz and a max of 12KB of ROM (so being able to easily change the code in the flash storage is an immediate advantage).
Running HTTP servers on tiny devices is hardly brand new - ok, this 2008 ATMega88 web server might be "cheating" using an Ethernet chip, but given the difference between this AVR's 8KB flash and 1 KB RAM there is loads of space on the Arm device to implement SLIP and do it all without extras.
Really don't want to demean this project - getting random boards into a programmable state is a pretty challenge in its own right and he did do all the necessary to get the response times down - but it is depressing to come across the line I've used for this comment's title, or variants on it, so often when MCU projects come up.
On the other hand, it is gratifying to know that there is (still) a large group of "non-programmers" that use/used original Arduino Uno boards (and equivalents) for art projects within 32 KB flash & 2 KB RAM when there are so many "pros" who can't imagine being able to work in such straitened circumstances.
To enable the drones to adapt to harsh environments. Assuming they last more than a week, what with the "enhanced hemostasis" giving them Deep Vein Thrombosis, strokes and heart blockages.
> Also left mysterious is how those SRBCs might get injected
Obviously, tubules extending from between the knuckles, duh (although these might get tangled with the Adamantium claws that other, equally sensible, DARPA proposals are working on).
But if you really want to know how well Smart Blood can turn out, have a look at
> So how old / new is Dirac Audio?
If only there was another way to find out, but I guess we will just have wait to see if anybody in the forum is able to remember when Dirac was created, as there are no records of anything outside personal memory.
Just one more problem caused by The Withering. Best call KENT.
We are okay, they can't issue work orders to the production staff to assemble anything.
Hmm, can anyone explain what we want with an order for an electric wheelchair, robot arm with a syringe and, what is this? "A golden cocoon within which to incubate my offspring". And, is that really Julie Christie's phone number?
Clearly, I know nothing about VMs - 'cos that sounds like it is opening up new and exciting ways for VMs to corrupt each other, which goes completely against the idea of having VMs in the first place (if you want memory sharing between processes then run them on the same machine in the first place, virtual or otherwise).
(Unless, is this just really bad naming for a.n.other feature, such as dynamically (re)issuing physical memory between VMs so they get just the amount they need, at the moment? Without actually transferring data between them, i.e. scrubbing the memory contents before allowing a VM to access it).
> Get a hammer and punch
Great if you want the exercise, or the therapy.
But for those who feel that an easier way must be possible, then an electric drill is up to the job*. Ok, you really ought to take the extra effort to clamp the drive down (also best to do that before taking a hammer and punch to it, btw); if you have a few drives to do then a simple wooden (ok, MDF offcut from B&Q bin, these days) jig will take fewer hammer blows to construct, be quick'n'easy to use - and, in the current climate of "refresh the fleet for Win'11" can probably be sold for a nice penny as a "specialist tool for safe destruction of hard drives".
PS
Drill into the cased unit, rather than taking out the screws; depending upon the material platters may shatter and keeping shards secure is best.
* Pillar drill even easier, but to be honest I don't anyone who has one if those - not in suburban England - but every bloke in the social circle has at least one electric hand drill. Actually, don't know that many with a hammer suitable for a punch, let alone the right sort of punch. Just hanging around with the wrong crowd, I guess.
> So, if a transition from 0.0002 to 0.0004 produces a significant effect, your, and my, perception of 'irrelevance' is, well, irrelevant.
Ah, you forget to take into account that these people are not only Observational Science Deniers, they are also Maths Deniers.
Catastrophe Theory - why, it is just a theory! And if we happen to go up a bit, we just have to go down a bit and it'll sort it itself out again.
Have you not seen the placards they like to wave: "Think Positive About Feedback"; "Hysteresis Doesn't Happen"
> why are some of us so unwilling/scared to acknowledge, and take responsibility for...
>> I am too old
"And simply don't give a shit about what happens after I'm in the ground"*
* Well, after the rest of his body follows his head
> locally, regionally, and globally
It is no good arguing against his arguments, the differences between geological time periods, the Miocene and now, are so big and impressive that he can just shrug off any claims that human activity could ever be capable of, say, paving over the local soakaways and causing nearby fields and homes (of wildlife as well as humans) to be flooded; of overusing fertilisers and causing long stretches of waterways to foul with the runoff; of over farming, with monoculture, and causing counties to turn into dustbowls.
As for global scale - pshaw! Why, even our biggest, most impressive explosions have only limited localised effects! So, your little box goes click more often, that doesn't mean anything.
They seriously believe that "Colossus: The Forbin Project"* is a documentary, or "Terminator"**, and the "AI" that China will build is going subjugate the human race, or wipe it from the face of the planet?
(Nope, tried a few times to come up with something other than "China might be sell more cloud time to access their AI than America manages to sell", but given nobody is running these LLMs at a profit, why is he worried? Except, of course, China "winning", then inevitably going bankrupt from doing so, means he won't have a bogey man to scare voters with hence it is the existence of his political career that is at risk. Clearly, that career is more important than global earning).
* other classic SF movies are available.
** see, told you there are other classic SF movies
Why not just go Full Octopus and use whichever set of limbs are useful for the moment?
The drummer from Little Mermaid had a friendly smile and wouldn't be at all threatening as it scuttled up the bookcase and across the ceiling bosses, carrying a teacup and sugarbowl.
More interestingly, Robert L. Forward suggested a fractally bifurcating design, nicknamed The Christmas Bush, in his novel "The Flight of the Dragonfly" (renamed "Rocheworld", how dull). Trickier(!) to actually build, especially with the way it separated into multiple parts (although those were very human-friendly, including a hair comb that could, among more serious uses, move to tidy and keep unruly locks neat all day). Think TARS etc from "Interstellar" but a lot prettier and more cylindrical.
Someone with the knowing could do worse than to set up a server that can be sent these sorts of claims - and it will then repost them to remind us when the date arrives, whether they come to pass (or don't, as may be).
Admittedly, it could be a bit - repetitive - with the antics of a certain person or two, but it would be entertaining nonetheless.
If I may suggest it be called Sherwood.
> Because the robot will work moreorless 24×7, won't require comfort or smoke breaks, demand pay rises...
If you're trying to make a case based upon costs, don't forget the flip side: when the servitor needs replacing, the robot is a capital cost, on top of the monthly repair and maintenance, whilst the new human is just getting pretty much the same as the last and no extras. Or you could lease, and lose control of your monthly budget to the whims of a third party.
Few humans are suddenly rendered automobile because the vendor has decided to deliberately pull support, close their servers just to force you to buy the new model (or because they went bankrupt, were bought out by a competitor or simply realised their biggest revenue stream was actually their packaging/logistics department).
Most humans come with standardised attachments; sometimes, there are a few inoperable items, but rarely do you encounter extra limbs. The next model of robot will come with, or without, whatever is deemed fashionable (or were available to the manufacturer in a cheap lot) and the uniform tabards will need to be remade, again.
> Humans invented the wheel
> Nature never invented the coaxial slip-ring
Don't place any bets on that.
Fun how we've always tried to keep claiming that Humans outdid Nature with regards to the Wheel. Nope. Nature has Wheels, many of them.
What we have is The Axle. And, in particular, the Heavy Load Bearing Axle.
Using axles, we've been able to make useful rotary devices at a large range of scales, from watchmaking to open cast mining machines, which have indeed surpassed what Nature has managed (so far, and in the ecosystems that we currently have available). But when we start to look downwards, we spotted the Fabulously Famous Flagellum under the visible light microscope, then all sorts of amazing machinery at the molecular level, which is where Nature absolutely rips us to shreds, cleverness-wise.
> when they should be starting with something simple
The "upstarts" are trying to roll out ludicrously, pointlessly, complicated products, for some weird reason.
Actual robotics firms have been in business for centuries*, gradually improving them as the techniques, materials and mass production of parts improve.
Aha, lookie there, the magic words "mass production", your clue to where you should look if you want to see successful robotics in everyday use. Well, there, and your washing machine; the old one, that went whirr and clunk rather than the one that beeps when it loses the WiFi connection. And anywhere that has a lifting device controlled by a yellow box with one big red button and a few smaller ones; the fewer the number of buttons compared to the number of motions available the "more of a robot it is".
* Yes, I said centuries. You do know about the looms, don't you? And, no, you do NOT need to have modern computers, or even any electronics at all, for it to be a robot, that is just reflecting the current (and not always necessary) trends: when all the robots are controlled by self-renewing organic artificial neurons (just keep the glucose tank topped up) they will be mocking you for claiming that anything using something as crude as "electronics" could ever be called a robot.
> interesting times quote originated in the counterweight continent.
It probably wasn't ever really a Chinese curse, but, um, *really* hoping you aren't suggesting that Pterry created the phrase - aside from anything else, if it wasn't already well known in English then we'd not recognise why he'd titled the book that way when seeing it on the Waterstones bookshelf!
You *do* know that NASA has been doing other stuff since Apollo finished?
Oh, and that the later Apollo missions did science, once they'd demonstrated the basic idea worked with 11?
Not forgetting that NASA was founded in 1958, so if it was purely about Apollo and they did those things before the end of *that* decade that would've been really impressive...
We can already see the turnstiles - haven't you noticed the increasing number of sites that suddenly stop and put up a page that "checks the security of your connection"* or that you are a human? The ones that get triggered if, for example, you are skipping through a catalogue (nope, that's not the one, next; closer, but no, next; oh for bleep's sake, not this tick box again)?
So far, I've not found anything trustworthy on how well these interruptions work - although they will slow down bot crawling, the same way that you use delays in logins to slow down attempts to brute force entry. Anybody got the data they'd like to share?
And talking of data, anything on user responses to this? If humans get pissed off and close the page, then your turnstiles are failing the other way.
* Stupid claim, nothing to do with security on either side; gain your customers' trust by lying about why the page is being presented.
True - which means that it is *way* past time to have had The Talk with management. If you started right now, you *might* find all the places that VBScript has crept, unnoticed, into all the company workflows*, but will you replace every single one before the Final Update** that stops them all working?***
* and don't forget that "workflow" not only covers that document automation system you got 17 contractors in for, back in 1998, and still boast about, but also means the 114 "just a quickie" scripts that were done as favours to PAs, as trivial fixes for the Sales team, as just something HR needed "only for one week as we get used to things", ...
** and company policy is tight, all the machines get all the updates, except for those two that are kept off the network and run the no-longer updated UI for the machine tools (see, we *have* thought it through and allow for special-case exemptions, we won't be caught out)
*** bugger, it was 116 quickies and nobody got paid: all the bank transfers went through without a hitch, but because those dozen lines of code failed to copy the exchange rates over, the multiplier was left at zero...
There is (apparently) a ready supply of Finns with the correct attitude, now we just have to find another one that is good with technology; THAT is the tricky bit.
So, are the people* this is aimed at smart enough? Or, as I asked, does Signal - or anyone else you want to rererence - provide a usable mechanism for the idiot 'phone user on the street?
It is all very well your being smug about this, but that is the very behaviour that leads to oh, so very many crap pieces of user experience.
* Note: *none* of this Signal etc stuff this is aimed to me, in case you hadn't guessed. So what I - or, for that matter, you - know what to do is functionally irrelevant.
And can only be unencrypted by the key that is safely & securely stored only on your phone.
Um.
PS as this got such a quick reaction, come on, how many are actually going to memorise a 64 char key or manage to otherwise back it and restore it; does Signal even provide an easy enough way for only an authorised person to do that for the key: "Signal stores it in an encrypted folder, backed up with a 64-character key stored on the device in case you need to restore access"
Was met by a sea of amazed looks on the faces of the gathered C-suite audience members, who were later seen puzzled by the live demonstration of making a tower from wooden cubes.
What a world to live in: not only is Gartner talking sense for once, which is newsworthy in itself, but it is only by repeating what every school child has been told since the invention of the pointy throwy stick.
But will anyone think of the cheaper alternative to taking tests? Just switch the damn AI off for a fortnight and work on the code (or whatever) "by hand"? Roll up the sleeves and dive in.
So many of us had to read E.M. Forster's "The Machine Stops" at school, now we know why; it wasn't just for the use of language...
PS is it deliberate that they named that product after the way one usually responds upon hearing what platform management want you to use?
"Make a database"
"Great, I'll just install MariaDB"
"NO! We demand SQL Server"
"Ssssss - MS" (holds up fingers in a cross, to ward off evio)
Because somebody had let slip that changing over would mean a rework, so they get a new version with Copilot stripped out and, with any luck, a long, long dev period before it reappears.
Devilish cunning, these SSMS survey respondents.
Get what you mean, but:
> been far more useful in human evolution
Pretty sure all the radioactives on the Voyagers went safely with them and won't affect our evolution.
Windows, on the other hand, is a toxic heap that has probably caused the premature removal of genes from the pool (via mechanisms such as self-defenestration: "No, no, I won't admin your office full of Win ME boxes, I'll take the less painful route - aaaaaaaahhhhhhhh").
We know an xkcd about that* citogenesis. Though your friend was being a bit more - deliberate - about it.
* There is always an xkcd...
> YMYL queries? Does that include 'How much pocket money do Americans get?' and 'How do I get rid of zits?' Not really life or death ('LoD'?).
Your particular choice of YMYL questions are, indeed, trivial, as well you know. But they aren't the only ones, are they? There are queries that people make which do have bigger consequences - so, taking your lead, from the totally trivial end of the bell curve to the totally idiotic end, how about "Should I withdraw all my cash and put it into memecoin?".
More realistically, closer to the middle of the curve, "How much can I transfer from X to Y in one financial year without going over the limits and being hit with taxes/excess charges/jail time for "structuring"? But phrased more innocently, such as "How much can I give my grandchildren this year?"
Get the wrong answers to that and you'll only find out months later when the taxes are deducted at source, or demands arrive at your door. Or the Feds.
Or "if I still send out twenty job applications a week, until one gets a response, can I volunteer at this shelter for an hour every evening without affecting my benefits?"
> 'Model collapse' is a little dramatic even if El Reg is journalism.
Model Collapse is the description used by the people researching how these models work - El Reg is just using the same term, to help you look up the papers that perform the analysis.
Your analogy with historic resources goes part-way along the correct lines, but only part way. Yes, daft LLM generated non-facts (aka hallucinations) will be picked up and repeated by the next generation of LLMs - and mutated again. Unlike (hopefully) historians looking to get back to only the most reliable form of the tale, the LLMs are taking everything without any critical analysis. Meanwhile, other branches of knowledge should have better self-correcting: even if an LLM confidently tells everybody that water boils at 32 degrees C at sea level, humans can just repeat the experiment (i.e. make a cup of tea) but the next generation of LLM can happily ingest that without thinking "nah, otherwise..." ('cos the LLM does not think, at all).
Also, unlike historians, the LLMs are also degrading in other ways - we've seen discussions over the last few days about spotting LLM output by excessive use of en- and em- dashes, and previous stories about how they are inordinately fond of certain phrases. Feeding these back into the training will only amplify such effects and, one mathematical catastrophe later, generation k LLMs degenerate into total gibberish.
It is just simple accumulation of errors: a positive feedback loop, which can only lead to the whole thing going up in flames. With the additional quirk that nobody really has any idea why these things are even as "good" as they are (at generating plausible gibberish): that is, you can't point at any arbitrary one of the individual numbers in the billions of nadans in the model and say what that one does. Or how to spot if the numbers are creeping up towards a point where *this* set of outputs is suddenly always going to be more likely than any other. At some point, after reinforcing their bad habits too much, LLMs fed on LLM output could quite happily become unable to answer any query without descending into a rambling discourse about the proper use of herring in Nordic cooking. Or just banana banana banana.
Oh, if only my keyboard had a "Windows key"*
* in fact, it does, as I was playing with the power toys thingie last year, but for the life of me can never remember which key I remapped** - Scroll Lock?
** Third World Problems - I have unused keys on the keyboard!
