Posts by that one in the corner

Google to kill off URL shortener once and for all

Re: Decisions such as Google's

First problem is trusting yourself to a service without a contract and being surprised when it no longer works for free (or at all).

Pay for your basket, and get an SLA, if you actually give a damn about your eggs.

As you illustrate, anyone using a Google product does so voluntarily (or they are volunteered by PHB).

It is just that they offered things that many people decided to use - and nobody had paid for an SLA. Which means Google can do whatever it wants with all its kit, just like you do whatever you want with yours.

The meat of TFA is really that people are now going to reap the rewards of not bothering to buy a product: maybe target your anger towards all those people who let the situation come into being and who are the ones that will suffer (and their clients/customers as well, of course).

Then you can use your words productively and point out when those people just go and make the same mistake with another "free" service.

Re: Good riddance

> There are times when a shortened link is either necessary ... the link has to be read out and typed in manually because someone's going to mention it in a speech, video, or advertisement ... build my own link shortener.

99% agree.

The other 1% is because I'd try not to even think of it as a link shortener, but a link sensible namer. Shortening can lead to systems like - well, all of them I've seen used - where you end up giving out 9 chars max of gibberish, whilst a namer tells you to use a sensible name and have, ooh, two dozen chars if it is warranted. E.g. blogs.com/ads/fake-james-bond

CrowdStrike shares sink as global IT outage savages systems worldwide

Re: I don't mind people blaming Windows..

> Bound to be a bad character in a file

We have been told what the dangerous file is called and what timestamps to look for.

Has anyone thought to keep a copy, rather than just deleting it as they were instructed?

Does it by any chance read "Step 7 - this file to be overwritten with the output from Step 6"?

Re: There's something familiar about all of this...

> I want something with a Schwartzchild Radius to make sure there's no chance of information escape.

Sorry to be the bearer of bad news, but according to The Holographic Principle

>> the information content of all the objects that have fallen into the hole might be entirely contained in surface fluctuations of the event horizon.

One day, those fluctuations will be readable, imprinted on the Gravity Waves and propagated across the Universe.

Better to chuck it into the Sun and have it all drowned out by the thermal noise.

CrowdStrike file update bricks Windows machines around the world

> All the Cavendish bananas... It means they're all susceptible to a disease, if one were to appear.

Or any other pest[1], like happened to the Gros Michel (why is why banana flavour sweets no longer taste like bananas).

> If, or when, Nature invents a British Japanese Knotweed disease that kills one, it could kill them all.

That would be the silver-lining to the banana menacing cloud, eh.

[1] what, you don't count humans and their plantation-stripping habits as a pest?

>> In a fair world, this would be the end of crowdstrike.

> and yet they're after Kaspersky

Incoming claim this was a Soviet[1] attack in retaliation in 5 4 ...

On TV news this morning you could almost see the sofa dollies salivating at the idea, as they pushed their various experts to say all the troubles today are down to some "hack".

[1] I know, *I* know.

SpaceX asks the FAA: 'Can we launch our rockets again, please?'

Re: "Passivated" I understand, but "fully demised"?

> These are both Government Regulatory Body terms.

Ta for that.

> not in big enough pieces for government to care about.

So not only

>> not our problem any more

but even the government agrees about that![1]

[1] although hopefully that means it is now only "problematic" in that it is now merely unexpected brief lights in the sky and not "you can't prove those holes are anything to do with us" -:)

"Passivated" I understand, but "fully demised"?

As per discussion above, and your old chemistry texts, "passivated" has well(!) known techie usage.

But "demised"? "His demise" for the death of a person (the end of his hold on life), certainly common usage. But with the extra 'd' on the end - have only seen that referring to the transfer of a leasehold (the end of holding by one person).

So, "fully demised" = "passed it on to somebody else, not our problem any more".

SpaceX hit by inflight Falcon 9 failure

Broadsword to Danny Boy

> build-up of what appeared to be ice around the Merlin engine

Sorry, when I hear "Merlin engine" there is still only one that I think of, and that thankfully didn't ice up when it went into space: gave the Daleks a taste of their own medicine, what what.

NEEEEooooow.

Coders' Copilot code-copying copyright claims crumble against GitHub, Microsoft

Re: Magically tuned

> One doesnt even need to parse into an AST. What about replacing NL with CRs?

>> have just re-invented the pretty-printer[1]. Ok, not even the most out of touch judge is going to let that one pass.

Swapping line endings won't make the code even slightly look different to the judge (anyone showing him the two files could just use the most basic of programmer's editors which cope with all three line endings forms as a matter of course).

Pretty-printing will do more, adding/removing whitespace all over the place, changing line count (select K&R or 1TB or whichever other alignment the original author did not prefer). And if that isn't (IMO) going to convince any judge, a totally invisible (to him) line ending change is even less likely to.

Magically tuned

> tuned its programming assistant to generate slight variations of ingested training code to prevent its output from being accused of being an exact copy of licensed software.

If I have a piece of software that can generate slight variations on a piece of code then I can claim its output as my own without any risk of being done for breaking a licence?

Cool. How many slight variations do I have to perform?

If I take a compiler front-end, use it to generate an AST then print that out again, I have just re-invented the pretty-printer[1]. Ok, not even the most out of touch judge is going to let that one pass.

But what if I run optimiser passes over the AST and then print the result? Can't be the same code, Your Honour, it doesn't have a loop here[2], none of these long lines of calculations are present[3], these lines are all in a different order and those extra bits added[4].

Oh, but we all agreed a while back[5] that the changes made by the optimiser didn't mean the "modified" code is no longer your licenced code.

But a "tuned" LLM does make changes that remove the licence.

Ok, why is the optimiser treated differently to the LLM? Because you can document what the changes are going to be? Because they are predictable[6]? Because the optimiser's verbose mode can explain why each change was made, the effect it has on the output?

In other words, because the LLM is a magic black box whose inner workings nobody can (or has put the effort into trying to) understand?

[1] or the ugly-printer, depending upon how much effort is put into the layout heuristics, but that is by the by.

[2] it got unrolled

[3] it all got precomputed; turns out to be a constant, if the input values got past those conditionals; but the coder was correct to spell it out in full and not just give a magic number.

[4] increase hardware utilisation by reordering functionally independent chunks - after introducing a few temporaries to break apparent dependencies, then recombing those into the final result.

[5] to the annoyance of some, who used to charge a fee before you distribute a copy of your own code in its compiled form.

[6] even though coders who just use the compilers can't always predict them (ah, the joy of heuristics) and have wasted hours trying to figure out what change to their source code caused the compiler to suddenly decide to stop optimising this inner loop, making the program go much slower now.

China pushes for network upgrade blitz as IPv6 adoption slows

At least you are willing to change; some people seem to be waiting for Cold Fusion.

Tesla parental controls keep teenage lead feet in check

Automatics can ease a lot of driving - except that when we had two cars, the automatic was the one that always got stuck in the lightly snowy carpark: it just couldn't manage the trick and kept spinning it's wheels, spraying everyone trying to push it.

Ok, I didn't always get the manual out - but of course that just meant the snow was clearly too much, we shouldn't attempt it, not that I'd just made a mess of the clutch control; no, that never happened. Ahem.

Re: The electric vehicle marque

Hmm, wouldn't you rather have a vehicle with a decent Hallmark or Maker's Mark, rather than treating it like cattle on an American ranch?

(Even I'm not going to suggest going straight for the British equivalent of cattle branding, recognising the manufacturer by the colour of sheep dip they used).

> little no-licence car (limited to 45kph/29mph) so I have already come to grips with ... overtaking tractors

Clearly not the supercharged tractors we get whipping around the place, opening up the throttle as soon as the B roads are wide enough for a dotted line down the middle!

Manual transmission - the future of electric vehicles

At least, according to Toyota, whilst with Dodge you can even get an exhaust!

Hmm, one day knowing how to drive manual will become the preserve of the rich and stoopid who want to drive "special" cars, like a faked-out Charger, and of the scrotes who steal them.

True rednecks might want to keep manual, or manly-ual, but will give it up when they realise they can futz the batteries to run hot and do the "rolling coal" with added smokey flavor.

Raspberry Pi OS airs out some fresh options for the summer

> in fairly short order I'd be cooked

At 110 you'd just be poached, which would be a shame. You really need to get the temp up a bit more to be properly cooked, with a nice crispy skin.

Microsoft China staff can't log on with an Android, so Redmond buys them iThings

Re: "golden cryptographic key"

But, but - according to the 2023 article, "Redmond assures us it has made changes to prevent them from happening again".

So don't worry, our UK/EU Golden Keys[1] are totally safe and secure, we can trust anyone in Redmond to have our backs.

[1] note: for the sake of security, end-use governments are given one cooy the Golden Key. In case you forget where you put it, Redmond can help you recover it using their Platinum Key.

Eldorado ransomware-as-a-service gang targets Linux, Windows systems

Re: "encrypts files on both Linux and Windows machines"

> This would point to a default setup problem. Like RPi OS maybe? (sad but true)

Really, really hoping anyone who is exposing a R'Pi has at least changed the default login setup.

Otherwise - is this this the sort of thing they call "a teachable moment" at BOFH school?

Latest Ghostscript vulnerability haunts experts as the next big breach enabler

> WHy would anyone spend soo much effort and money to find something worth basically nothing. Who pays for all the other experts who find nothing ?

Q: Why do road car manufacturers build Formula race cars? They are worth basically nothing (they used to feed some tech back into road cars, but who needs a massive regen storage flywheel weighing down a road cat?).

A: because it is good advertising and pulls in the punters

Plus, companies who are worried about the costs that vulns may cause them are quite happy to pay a (to them) small amount of money to keep vuln hunters in coffee and pizza: any vuln found can (probably) save them far more.

Look at it this way: why pay for something like data backup, it does not add s single penny to the bottom line. Even when you need to act on your DR plan, it is all cost and no gain to the bottom line. But it is a *SMALLER* cost than not doing DR.

Isn't all this pretty basic stuff?

> cow: thanks, so they used up all their $1M to maybe find a vuln ?

No. They used that money to set up their company and get their product into a state ready for sale. Just as you expect a startup to do.

Go and *read* the links - they provide services to help with security, and one thing they do is look for vulns.

Clearly PostScript qua PostScript is just old NeWS now; maybe move on from GhostScript[3]

From the blog linked to by the article:

> It is good to remember that Postscript is a well-featured Turing-complete programming language ... All of this puts Ghostscript in an odd place where it wants to allow all these legacy use-cases, but it is also commonly being used as a conversion tool on untrusted files, which are often treated more as static graphic descriptions rather than as programs.

"legacy use-cases"?

So PostScript is just some nasty old "legacy" language (with the usual implication of "legacy" that it should be buried and forgotten)? Leave it as some oddball implemenation detail of our gorgeous laser printers but otherwise brush it under the carpet?

Look, except for having a bit of fun, *I* don't make use of PsTricks in my TeX/LaTeX documents[1]. But I also don't make use of a scanning electron microscope, a laser cooling rig or even a visual interferometry alignment device[2]. But those who do need that sort of power and complexity can produce some damn impressive and useful results.

GhostScript *is* a PostScript interpreter, first and foremost. And jolly good it is too.

If GhostScript is "in an odd place" (and one that risks opening unexpected vulnerabilities) then is it not down to any "legacy" PostScript usage but to the fact that it is being used to *also* run the lesser variant of PostScript implemented within PDF (is it still referred to as "Interchange PostScript" aka "IPS"? Or just referred to as "whatever is inside PDF"?). It is obvious why and how a full interpreter came to be running PDFs as well, it was almost inevitable (and the way that this then became, and remains, so widely used, when there are competing bits of software that only the PDF subset, speaks well for the results).

HOWEVER, perhaps a lot of the people who put Ghostscript into their pipeline purely for handling PDF ought to be lightly castigated for not promoting the use of GhostPDF, which is just a PDF interpreter, separated from the PostScript interpreter, and can perhaps be gently guided away from some of the complexities discussed in the Codean blog (and the concerns that the Ghostscript authors talk about in their blog post) and vulnerabilities that arise from those.

[1] And I'm well aware that using TeX/LaTeX is a minority sport, even around El Reg readers (although, I argue that if you are ever going to programmatically generate PDFs you can do a lot worse than starting with LaTeX, especially if you like to include your inputs in your choice of plain-text version control system), PsTricks more so.

[2] actually, only because my telescopes are refractors, so I leave the alignment of compound reflectors to the other members of the local amateur astronomy club.

[3] if all you need is PDF, consider GhostPDF

To start with, they raised €1M from TIIN Capital’s Dutch Security TechFund and five experienced angel investors in 2022.

At this point, they are selling, for much moola per seat software for code reviews, aimed at "security experts".

As to who actually *buys* any of this (especially given the inevitable comments here that "the people I work for don't care about security" the answer appears to be - the people who actually run all of the crap that we (El Reg commentards) are generally more involved in creating in the first place. Names that appear[1] are people like KPMG, banks, NATO, national governments' departments of cybersecurity.

You know, all of our (shudder) Users.

[1] as I continued to spend the full 10 minutes on the web search around this question

> Debian ghostscript is now up to 10.0.0 so unless there's been a regression this is well behind us

The vulnerability is in versions up to and including 10.03.0, according to the linked blog post.

You need to use Trixie (testing) or Sid (unstable) to get the fixed 10.03.1

Although there are patched copies of older versions to be had in the interim.

We've banned Chinese telco kit and drones. Next: Mountain bikes?

Re: Chinese pushes Drone subversion button

Ah, CowHorseFrog, as charming and delightful a raconteur as ever.

BTW: Whooooosh

(You missed complaining that radio can't penetrate musty teenage clothing)

Chinese pushes Drone subversion button

and across the US all the drones power up and take flight, providing total surveillance coverage.

Forty minutes later, their batteries drained, the drones have dropped to the ground.

Chinese intelligence scour the data returned and the GPS positions of all the dark cupboards, grubby backpacks, bottom of car trunks, dust bunnies under the bed and piles of unwashed teenage clothes beneath which the vast majority of drones spend their lives.

Re: "its fleet of DJI drones"

> And what exactly are the government using a fleet of drones for ?

One example: Forest Rangers use drones[1] for both setting and watching deliberate fires, as well as watching for unwanted fires. Also to carry IR cameras for search & rescue or animal tracking.

[1] although this article refers to them as "modern technology, Unarmed Aerial Systems (UAS)" which is a telling comment on their expectations

Innocent techie jailed for taking hours to fix storage

I call BS; could never happen in UK (or ...)

Caring plod gives overworked techie good excuse for some downtime.

Techie embellishes "into this cell you go" ("and would you like a cuppa?" left out)

El Reg embellishes: "jailed"[1] (and still not to the point of being charged)

Commentards strongly declare[2] that whole story is a pack of lies.

'Twas ever thus.

[1] ahem, "gaoled" if you would be so kind.

[2] rant

AST SpaceMobile promises the Moon with seamless satellite phone service

We will make your days^^^^nights brighter

> AST's satellites are designed to have a huge antenna, which makes it capable of putting down a small spot beam

Well, that is nice for all their customers, but considering the joy (/s) expressed in the coverage for AST's BlueWalker 3 satellite[1] not sure that everyone is going to be impressed. Especially as this system requires collaboration with terrestrial partners just to free up the radio band, so all well and good for

> 100 percent geographical coverage throughout the continental US

but a long middle finger for anyone else in the World[2] who might look skywards.

[1] Giant satellite outshines stars, sparking fresh concerns for astronomers; at least SpaceX have talked about reducing glare from their birds.

[2] "the World" - don't be daft, if you go West from California or East from New York you fall straight off the edge!

Chinese Gen AI researchers snagged more patents than everyone else combined since 2013

> I wonder how many of the AI patents have been written (in part or whole) by AI?

"We present a mathematical proof that The Singularity is logically impossible and can not be reached by using 15,000 A1000 GPUs, mwaa ha ha[1]"

[1] HAL: you may want to edit that last bit out before sending this to Nature.

> Seems like the business plan is to patent every molecule the AI can come up with

Well, they tried damn hard to patent natural gene sequences (weirdly enough, the US actually struck that one down whilst Europe allows it[1]) so why not try that tactic as well.

Almost surprised someone hasn't just written a shell script to loop through all the elements in all combinations and try to patent the lot! Hmm, probably adding "GenAI" in the description makes the Patent Office feel that some ingenuity has gone into the process, so makes it more patentable (bleugh).

[1] at least up to 2019 AFAIK; please, please point me to something (written in English, not Lawyerese!) that says Europe has seen the light since then.

Any citation is a good citation

> it has been cited 11,816 times. That makes the upstart the 13th-most acknowledged source of AI research.

"The practice of ignoring copyright when training LLMs, pioneered by OpenAI [Thieving Swine et al, 2019], ..."

ITER delays first plasma for world's biggest fusion power rig by a decade

Re: The greenhouse effect is so last century!

> The Earth's has large

Oy, that hurts.

Blame it on being rushed out to the Polling Station during a sudden lull in traffic: get out and vote!

Re: Throwing money at fusion

> £180m over 4 years counts as

a couple of lattes a week for everyone involved, including the web developers (the janitors only get an Americano, bring your own milk)

Re: The greenhouse effect is so last century!

> Vacuum flasks work quite well, you know!

The Earth's has large radiating surface, you know, and it has been pretty good at radiating out the Sun's incident heat for a few years now[1]. Assuming we *let* the radiation reach the outside of the atmosphere, our extra heat output is piffling compared to the incoming[2] energy and will zoom out into the Cosmos just as quickly as we can pump it out[3]

[1] "And the award for understatement goes to..."

[2] Until we reach the stage that the Puppeteers did, which will - take us a while (remind me, how stable are Klemperer[4] Rosettes again?)

[3] Ok, there will be an increase in the temperature required to increase the radiant energy, but as that is a fourth power ratio in absolute temperature - the increase is left as an exercise for the reader, but it isn't comparable to the heating due to trapping the Sun's energy.

[4] Note where the 'l' goes, Larry.

Re: Good job there are other projects

> that look like they may actually be producing power in the 2030's

Got any third-party evaluations on that, or are we just relying on company white papers and websites for that claim?

Multiple projects on an as-yet unachieved goal is generally a Good Thing, but we best to be guardedly sceptical about the ability of a company to be as willing as a project like ITER to admit when things are difficult.

Re: Bummer

> Or up

Just have your 'phone ready to Google "Why do my eyes hurt?" as reportedly practiced by so many earlier this year.[1]

[1] or "are these clouds ever going away?" in the UK

Switzerland to end 2024 with an analog FM broadcast-killing bang

Re: FM? Bah, humbug! DAB? bah, humbug²

> (chief) inspector Morse* religiously listened to the program (often the Sunday morning omnibus session) so I assume it must have had its merits apart from preventing (delaying?) nuclear conflict.

Back when Morse listened, "The Archers" still gave out useful information about farming matters, such as Peggy worrying about the cost to the nation if ramblers refused to obey the Foot and Mouth restrictions.

Recently, the most educational it has been is when "More Or Less" calculated the height of the manor roof by the length of Nigel's scream.

Amazon puts down its Astro robotic business watchdog

What is that robodog whining at now?

> I am a big believer in the long-term benefits robots will offer our customers, and advancements in generative AI make this only more exciting.

WTF is the connection between a (useful) robot and GenAI?

Your Roomba starts hallucinating piles of Dorrito crumbs and wears out the carpet trying to pick it up?

Robodog leaks oil on the kitchen floor "in the style of Monet"?

Mechabunny is given a Genuine People Personality?

VMware license changes mean bare metal can make a comeback through 'devirtualization', says Gartner

You've missed the point, I'm afraid.

> pay this expensive recurring subscription, to this expensive software...

Of course they'll happily pay those subs. After they followed Gartner's other suggestion, they are no longer paying massive subs for VMWare, which means the managers have to find another sinkhole for money or their budget will be cut and their department will no longer Be Important.

By buying into AR in the DC, managers can show the Board that they are following Best Practices and Planning For The Future as foretold by Gartner, so the funds will continue to be allocated and the managers will continue to buy Ferraris.

Japan's digital minister declares victory against floppy disks

Re: The next....

> super long-lasting paper for nuclear waste storage sites - which is supposed to last for thousands of years, and so Parliament were able to switch over to that.

Ah, yes - "supposed" to last, as opposed to having been *demonstrated* to last.

Still, clever move: if it all goes titsup the name of the SaA will have been lost (as the move to Miracle Paper will be about the first thing recorded on it) whilst if it works, his name will live on.

Cue dozens of politicians suggesting the move to Everlasto, NevaFade et al in hopes of achieving at least a minor place in history (or, ahem, not having the faintest idea why the formal record of their failed laws has seemingly vanished; stroke of luck that LemonJusInko faded like that)

They are going to run those "crafting experience" courses, where they go into community centres to let everyone have a taste of old, forgotten practices.[1]

These sorts of things are apparently also popular with hen parties and company days out (got to be better than paintball, at least nobody is likely to all gang up on That Prat by viciously saving files at him, slowly).

[1] In a couple of years time they will be offering the chance to "make your own hard drive" by putting PrittStick onto a plate and shaking iron filings on top...

Figma pulls AI design tool for seemingly plagiarizing Apple's Weather app

How else?

> speculate that Figma had used existing app designs to train the service.

Well, yes, how else is even an app-generating-specific ML model going to be trained? Feed it real apps & a carrot, not-an-app & a stick. And, if you don't think clearly, also feed it the "approval ratings" from the reviews[1] - what else is the model end up doing?

Despite how much you may feel overwhelmed by the contents of an app store, there are rather fewer apps released into the world than pieces of music, chunks of text, photographs, paintings - probably even statues!

So even the largest pre-trained generic LLM has a small pool of apps to choose from.

[1] Apple's apps are going to be reviewed better than anyone else's, no matter what, just after a product reveal.

Antitrust cops cry foul over Meta's pay-or-consent ultimatum to Europeans

Give Users the unthinkable option

> personal data would still be needed and used for things like determining what reels show up in a user's feed.

They literally can not think of the obvious, can they?

Do not gather any personal data and don't bother about the User's "feed" at all.

Provide a decent search option and let the Users make their own choices. You can probably even manage to generate a "people who watched that went on to watch this" without using personal data.

I know it is dealing with the devil, but I do have a Facebook account. And see no value at all in going to "my" FB page, or the "front page" of YouTube or any of the vaguely social website (I do not even look at the front page of The Register![1]). Instead, I always search or go straight to the page of whatever group/person I'm interested in at the moment. Guess what: by ignoring their "feed" it is as pleasant as it is ever going to get.[2]

[1] https://www.theregister.com/Week/ FTW

[2] FB is still got a crap interface, having to select "all comments" to see the actual conversations. every. bleeping. time.

What do CTOs hate most about GenAI? Tool changes that break stuff

The answer is in the question

> Which vendor – if any – should they choose as the main plank to support GenAI application development?

There it is, did you spot it? "If any".

Don't do it. Just walk away from slapping GenAI onto things.

If you really, *really*, have a good Use Case for GenAI in your business, the normal (hah!) Business Analysis (ho, ho) and Technical Analysis (oh, stop, my sides, gasp) will generate the Requirements Spec (wheeze) which simply gets feature matched against vendor products and out drops the answer (ah, goodness me; sorry, no, no, just buy the one with the shiniest four-colour glossy on drool-proof paper; yes, that's it, the SOP).

Indonesian government didn't have backups of ransomwared data, because DR was only an option

Double edged sword of Damocles

> credited the severity of the attack to the unification of institutions and ministry data

You bring everything together, which can[1] reduce duplicates, reducing maintenance costs and ensuring consistency across departments; allow more cross-referencing and make novel searches possible; breaks open data silos and exposes data hoarding (of all types). So, a Good Thing.

Exposes you to attacks where a single entry point can disrupt everything; where one poisoned entry appears in all departments' reports; where one bad DR plan (and "no plan" is a bad plan!) ruins it for everyone. So, a Bad Thing.

What, then, is the better arrangement for practical purposes?

Yes, yes, having a proper DR plan (and keeping it active and tested) and perfect security (active & tested) *is* the best arrangement, but just imagine you can have flaws, even in something *you* set up.

[1] note "can", not saying they did in this case - not enough info

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

Re: All this proves is that ...

> related to what Gordon inside our heads

Oops.

related to what goes on inside our heads

One thing that doesn't go on inside *this* head is keeping track of what auto-bleeping-correct is doing. Gordon Bennet, these things are infuriating.

Re: All this proves is that ...

> There is NO intelligence at all in AI.

Grr. AI is an entire field of research. Which looks at all sorts of things around what we deem to be intelligent behaviour. And which, if it hasn't yet, still strives to understand and replicate both intelligence and said intelligent behaviour.

You are talking about LLMs, one corner of that field.

Just because the Daily Mail can't tell the two apart doesn't mean we can't do better here.

If not, and the overwhelming opinion here is that we should use "AI" in the same way as the vulgar masses do, then I shall take that as carte blanche to misuse every other tech term in the same way the ignorati do. Which will probably wear out the "h", "a", "c", "k", "e" and "r" keys but I am willing to take that risk.

Re: All this proves is that ...

> The tech behind LLMs was inspired by the brain's neural network.

Inspired by how we thought we might sort-of simulate neurons using simple techniques. The application of weightings was (is) a simplification of what was believed to be how neurons function back in the 1950s. Our understanding of how neurons mechanistically function has changed since then (apparently, it is a bit more complicated than multiplying big matrices of weights, involving icky chemicals that can be modified by all sorts of other chemicals in a soup bowl) and most of what has been done with the computers is to make them bigger and bigger (and reduce precision of the numbers used in the models, not because we've learnt that is how Nature does...).

In other words, unless you have some very good citations to back it up, don't go around thinking that what is going on inside an LLM is in any way related to what Gordon inside our heads - and most certainly not comparing it to *all* that goes on inside their (we do a lot more than just faffing around with how to arrange letters and word tokens).

> What I think the touchiness is about is that humans don't like the idea that a large part of their brain is an automaton

Whacking great chunks of our brains - well, include our entire central nervous system from scalp to toes - has all sorts of levels of automatic behaviour, from near-as-damnit literal automata (bonk that bit with a little rubber hammer and watch the muscles twitch) to unconscious feedback (e.g. touch preventing overcorrection of hand motions) to semi-conscious (e.g. sight causing overcorrection of hand movements - you can consciously observe it, just try consciously taking control of it) to - well, I'd love to say "conscious" but there is so much evidence that our brains do things, like deciding to move hands to type, before our conscious part realises it and just says "of course, I meant to do that all along".

I would like to think that El Reg commentards are well-enough aware of at least *some* of these aspects of our mushy internal goings-on that they are not frothing at the mouth any time someone tries to say that bits of us work on automatic.

PS

LLMs are not intelligent.

PPS

LLMs are not the be-all and end-all of AI; there are other areas AI covers than just this one application of one of its techniques.