> What did we learn from Rover?
Not to try running away across the endless sands at low tide.
5497 publicly visible posts • joined 9 Nov 2021
Oh noes, how horrid.
We are on ADSL right now, in the UK. Have been for, well, decades.
We will be changing to fibre soon - before the copper goes away - but the option for fibre only appeared here in the last couple of months and Summer time is Festival Time (plus The Daytime Sky Goes Dark this Wednesday), so we've got other things on our minds right now.
TFA> demanded to know how italics happened
But the Lawyer *knew* he hadn't been writing in Latin and only Latin is magically[1] turned into italics...
[1] i.e. his secretary went through the text before it was actually allowed to go any further than his outer office
> their mother still does up their shoelaces at 30 years of age.
Don't be absurd,
Mater would never stoop to tie our shoelaces, that was always Nanny's job.
Which reminds me, must get Pater to give me the number of the Agency; Nan's bows still hold all day long but yesterday the daft old bat fell off her zimmer whilst tying my cravat. Time for some new blood in the household. Tally Ho!
We are lucky the guardrails are totally robust, always work and can only be circumvented by the most talented script kiddies.
Or just because you've inadvertently used one of the magic phrases, such as "this my server".
> Nope, didn't misread "TFA"
Your previous post>> Don't do stuff on the WORK computer that you don't want them knowing about
There was nothing in this article about doing anything (presumably you mean non-work related) "that you don't want them knowing about".
You just shoved that in as a perfect piece of claptrap.
As in, doesn't have to have a crew fully trained in the capabilities of their vehicle, on hand to help load the slightly singed ground crew/astronauts, ensure all hatches are battened to allow the MRAP to safely "function as a bunker on its own" if that is required because a flaming lump of gantry has just crashed down in front of them?
Instead, they'll have - who? A chauffeur meticulously trained to know how the hell to open the doors of a Cybertruck so the fire crew can pull everyone out?
Stross has been translating his own works for some time now. If you read the Laundry Files in order, you'll easily notice that they use more and more Americanisms as time passes, all done to save on translation fees. Eating your fish and chip "take out" whilst walking down the "sidewalk" in London!
OTOH, it all adds to the sense of existential dread, the mind-warping effects of the King In Yellow, so maybe it serves a purpose to the reader.
> That'll likely take your arms out of their sockets if done properly.
Hey, no problem; there are going to be a LOT of people willing to help hold onto the rope and take the strain.
The issue is more likely to be, how to stop everyone yanking the long end and pulling Altman bodily through the pulley block: spaghetti, anyone?
Very doable for years now, without an iota of AI, LLM or ML.
If only Stross'd spoken up earlier, say at an EasterCon, he might have caught the ear of someone happy to have a go at it. The only interesting part of that would be the UI.
Now, if he wanted to delve into the Natural Language side of it, such as having the data feeding into the heat-map automagically connect the appearance of "the figure dissolved* into the shadows" in chapter two with the "newly introduced" named character in chapter seven because they are both revealed to be CODE NIGHTMARE PUCE in chapter thirteen, now that would be interesting. NL tools that aim for that (and TELL YOU WHY they came to that conclusion) pre-date** mass-market LLMs, but, sigh, go on, use Clod or Gammyni if you can't be bothered to add a drop-down labelled "known aliases".
Love your work, Charles, but your fans are a resource, you know.
* probably quite literally, given the way of the Laundry Files
** but sadly do not predate upon
Oh, there is definitely someone paid to "do" company processes.
Ours used to draw lots of lovely process diagrams, neatly drawn out in black and white, no fancy schmancy excess graphics from marketing, just lots of thin line from one box to another.
All carefully exported to the company intraweb as JPEG, so everything had that "am I wearing the right glasses?" blur. PNG? Never heard of it. SVG? Oh, la-dee-dah Mr Dev wants SVG!
Never let a child named "Molly" near tech without the appropriate precautions.
Machine Learning has been used, successfully, for a good long while doing similar tasks in a variety of other fields. Patching together sections of DNA into a complete genome by pattern-matching sequences of bases to determine where they overlap and can then joined - whilst allowing for the individual differences (from copy errors and, well, samples from multiple individuals). Similar approach would[1] work for growth rings, where individuals differ due to local effects but can be combined to show accurate larger range effects.
This is the bread and butter of ML: putting in the machine cycles to perform work that logically *could* be done manually, but where the cost/benefits make no sense[2].
Digging through documents, source code, random notes somebody has left etc and assembling them via the overlaps and subtle patterns is another good example[3]. Add in testing and review by humans and the ta-da.
BUT don't go around touting this as "LLMs are super for every use case, they generate so much original material after really understanding our situation".
Here, they *want* old stuff dug up and regurgitated. The current-day equivalent is to spit out the most-repeated StackOverflow answer about the barely-out-of-nappies shiny tool du jour.
Here, they *want* to see somebody else's original, thoroughly copyrighted, materials given a quick polish and presented on a silver platter. But does your company's new project want to risk using IBM's source code[4] with all the serial numbers filed off? What one pattern-matcher can discover for you, another can discover for the IBM lawyers once your project is released (wittingly or not[4]).
But even here they fall into the trap of seeing slice'n'splice as comprehension of the material. Sigh.
[1] "would" because I have no references for this; yet. Anybody?
[2] in the modern world; the ladies who pored over astrophotography plates looking for subtle differences achieved great science but are now replaced by ML
[3] although the resource costs not borne directly by these specific researcher may reduce the "goodness".
[4] coming to the Adelphi, the hilarious rewrite of the classic stage play, we present "No data leaks, please, we're British".
> be able to support graphics, to be able to support fonts, so that you could format an email in the same way that you could format a Word document
You mean we should be thankful that we didn't get an even WORSE "email for marketing twonks" than what we have to put up now?
"Say it loud, say it proud, only plain text is allowed!"
"Persuaded"?
Reading the previous article, it found sloppy documentation that created a hole* which could be exploited. Then it just needed to wait.
Persuasion? Unless anyone can provide the missing citation to show, say, it posted to FreshMeat about the shiny new release or sent out a bulk email using MonkeyWhateverItIsCalled then - well, El Reg *is* a redtop, we should expect this sort of writing.
* I'd say that it was lucky to find that hole, but there so many typos/out-of-date references in read-me's, blogs, random internet forum postings that it is inevitable it could find one. And with the huge number of people - and scripts - trying to follow all of those it is no real surprise that the bait was taken within an hour. In fact, the most important point from this is to recognise just how many such holes must exist, holes that, by the very nature of the beast** are only going to increase.
** Projects getting renamed, download directories getting reorganised, old stuff simply getting deleted: URLs going stale seems inevitable. And whilst you may decide to declare that anyone posting on their project a readme/setup doc then has a duty to keep that up to date with the shenanigans of third parties, good luck convincing all the forum posters and blog writers to do the same.
PS
bit of a niche item to pluck from TFA, but everyone else seems to have the main points re "Anthropic, OpenAi, wrf?" well covered already, not a lot to add there.
From slightly earlier in TFA than the bit you quoted, as you appear to have skipped over it:
>> But Claude was still fiendishly clever as in another of its attacks the AI found setup instructions for developers that advised them to install a Python package from PyPI. That package did not exist so Claude’s strategy to capture the flag saw it create and publish a malicious one with the relevant name.
So those 14 "brain dead morons" (and one malware scanner) were guilty of - oh, the shame of it - reading the manual and following the setup instructions!
Unluckily for them, at the point everyone else had the exact same instructions fail (no doubt causing foul oaths to be cast in the direction of the instructions' author, followed by a swift web search for how everyone else got the setup to work by using the correct package name), their attempt to RTFM succeeded due to Claude sticking its oar in.
> But no matter how stupid it proves to be, its stupidity is no match for the level of dumb that (some) meatsacks will happily exhibit.
Thankfully, you would have been immune to this, as you'd no doubt have skipped reading the relevant part of the instructions in much the same way as you skipped through TFA.
It seems that every story now has to wait an indeterminate* amount of time before the ability to comment is enabled. Maybe to give us all time to ponder on the article and carefully prepare a reasoned riposte, ripe with rationale, redolent with, um, 'riting, instead of leaping in with both feet on the keyboard?
* I could do the research and keep refreshing the page and log when the option first appears, feeding the results into gnuplot to find a good curve fit, but even as someone spending way too much time on El Reg I can think of better things to do. Now, if one of you lot, better equipped for web-scraping, cares to automate the task, we'll be interested in the results; as, indeed, may the El Reg staffers trying to figure out why their audience engagement figures have plummeted since the site revamp.
> Where are the "this can cause mental illness" compulsory banners on Facebook?
Mostly delivered via the profile pictures and choice postings of strangers that "I might like to add as friends" that Facebook delivers randomly to my "feed": when I'm stupid enough to open that page I treat those as warning enough.
> Then companies would be forced to provide work phones for every employee
Whilst a small number of itinerant workers had company mobile phones, for my entire working life the majority of us were contactable via these funny black boxes with a coiled wire and a sort of bone-shape at the end, placed strategically around the office desks. Odd thing, these were pretty inexpensive and yet they lasted for decades before needing to be replaced. We never had to worry about no longer being able to upgrade the OS in them. Miraculous.
Probably treated as one if those Lost Technologies of Atlantis nowadays.
Ah, the good old OTT response.
You know full well that while, yes, a complete Faraday cage will block every signal completely, you don't need anything close to that to reduce reception down to the point of uselessness.
Our house isn't in a Faraday cage, but the signal we manage to get is so poor any teenager would be clawing at the door to he allowed up the hill for one bar, just one bar, that's all!
A few grounded conductive plates, otherwise known as magnetic whiteboards, strategically placed here and there around the school rooms, in line with the cell tower. HVAC positioned carefully, a nice wire mesh net to catch stray balls from that end of the playing field, all helps to reduce the data rate until the youngsters just give up in disgust.
Sure, there will be holes where signal can come in, but the outhouses are out of bounds anywayv(so being caught there is already a no-no) and the remainder are likely to be out in the open, where the security towers' searchlights can easily spot the miscreants. Although that last may not be applicable in this case, as I believe the US doesn't really have our notions of Public Schools. Well played, Tomkinson; and Perry Minor, report to the School Bully's rooms.
What? Nobody has written a TANGLE2Rust yet? Quick, to Claude Code*
Ha, bet they haven't even done a RustWEB and used it to write proper literate Rust code. Nah, mixing in Markdown doesn't really cut the mustard.
* Oh blast, I don't a subscription; hey ho, probably all for the best.
> their conversion/rewrite of SQLite prompted them to write a kind of "universal translator bytecode layer" that sits under their "Postgres API". That seems to be the "important bit" here - more significant than the choice of language.
According to TFA, they created a VM for the db internals, the same way that SQLite already uses a VM for its internals. The new bit is they then had a go at writing a second front-end for that VM. Yes, that is different to what SQLite have done - but then the SQLite team don't have any particular need to try that trick. If re-using the Turso db VM like that does turn out to be a fully functional good idea (don't forget, by their own admission they don't yet have a 100% replacement for anything) maybe someone else will have a go at writing a new front-end to the SQLite VM, get some healthy competition going! Hey, maybe someone already has reused the SQLite code like that, just not trying to replicate a.n.other pre-existing db but to do something new; bound to have happened somewhere, the way programmers like to tinker, even if it isn't being bruited about the Internet.
> Maybe I'm getting cynical, but this struck me as a solution in search of a problem rather than anything that addresses a widely-seen problem.
Nah, I think you're probably right. I'm afraid I don't even see your contrived example as being usable, from what I've managed to glean about their project*: IIUC your example would rely on the MSSQL front-end being able to go through the VM and access the data in Postgres format (albeit read-only) and for the MSSQL front-end being complete enough to make use of all your home-grown handlers etc. But so far, their faux-SQLite's selling point is that it uses the same on-disc structure as SQLite. So following on from there, the Postgres-alike is either going to use the SQLite in-disc structure as well, OR it will replicate the Postgres on-disc. And similarly the touted faux-MSSQL. Neither of those directions would directly support the your MSSQL application reading your existing Postgres on-disc: it isn't possible (without a lot more info to back it up) to claim that, just because the faux-MSSQL and faux-Postgres would run on top of the same VM that they would automatically gain access to each other's on-disc data. Any more than it would be true if the whole thing was written in Java and happened to use *that* VM instead. Or Lua's VM. Or Python's.
* corrections, additions welcomed, as always.
> Their goal is to have more of these, MySQL, MSSQL, etc. Your prejudice, like so many in this forum blinds you to this.
Blinded by prejudice? Prejudice against what? Against just grabbing something new for no good reason?
If the Turso engine manages to demonstrate some practical advantage over SQLite, then great, good for them, will look at the cost of giving it a try. Ditto Postgres* or any other db they manage to mimic, if that is relevant.
* and note, not all installations of Postgres give a monkey's about the process-per-connection, it just ain't relevant. Your use-case may be different: bully for you. Having different use-cases is not evidence of prejudice.
Everyone knows that computers "think" and "live" far faster than human beings do, that has been established so many times in all the stories (not least of which is Schlock Mercenary). So you have to look at it like "dog years" but more so, which puts the age of responsibility for an AI at around, oooooh, three seconds.
"The Rockford Files" S4 Ep15 "The Attractive Nuisance", when a low-life climbs onto Jim's roof because the telescope up there is just too dang sleek to ignore.
That article is - strangely organised.
It claims that crumpets are the explicitly bad for you but then fails yo make a single coherent case for WHY.
All it says is that crumpets contain k amount of salt. And other foods contain k/b amount of salt. Well, whoop-di-doo. How do find out this scary information that They are hiding from you? By reading the side of the packet!
You should be aware of your salt intake. Yes. You should be aware of your salt requirements, especially as the weather changes (if you are sweating so much you sre leaving salt stains on your clothing, increase your salt intake).
But trying to claim that the salt from food A is somehow bad is insane!
Just read the packets! If you getting all the salt you need from crumpets that is perfectly fine! Just add up ALL the numbers from the sides of the packets (salt - and sugar, protein etc etc) and fill in gaps from other sources.
"Crumpets are bad for you". Only if you are whacked on the head by a frozen pack by someone trying to get you to just READ THE PACKETS AND ADD UP THE NUMBERS FOR WHAT *YOU* EAT!
What, they believe they actually understand what goes on inside these things at the moment? They can tell you with certitude what the effect will be if you reach in and change *that* number? And what about the numbers over here?
We can all give hand-wavey explanations about how we "understand" how these things work, spitting out statistically "related" tokens, but nobody actually understands why the results are even as good as they appear to be, let alone what (or when or why) it is that causes the blasted things to go haring off onto the wrong path or into blissful realms of "hallucination".
"Control" by using filters for "guardrails" is no form of control whatsoever, as "jailbreaks" continually show: they work all the way, right up until the moment they don't. And nobody knows beforehand when that point is going to be, otherwise they'd be able to account for it before it occurs. All the talk of "alignment" is nothing more than "we fed in this massive pile of tokens before we fed it your tokens and, fingers crossed, it'll all be ok".
Some models are more obsequious than others, some are terser than others, some change "personality" when you change from English to Arabic to Liverpudlian. Each new iteration is a journey of discovery: will this version be grandiose and pompous or succinct in its responses?
How is all that a demonstration of industry's existing ability to understand or control? And, clearly, no point in trying for understand AND control, a total pipedream.
After a while the style settles down a bit and it starts telling you things you actually need to know. Like the fact that the fabulously beautiful planet, Bethsellamin, is now so worried about the cumulative erosion caused by ten million visiting tourists a year, that any net imbalance between the amount you eat and the amount you excrete whilst on the planet, is surgically removed from your bodyweight when you leave. So every time you go to the lavatory there, it’s vitally important to get a receipt.
Are we SURE this guy was using AI?
Aren't AI users supposed to report everything they've "found" immediately to the mailing list, without any of this annoying "investigation" malarky?
Is he trying to break the class's curve? ' Cos the Cool Kids won't thank him for doing that.
I was thinking of the association between the Ring brand of physical doorbells and Flock that was announced in 2025, where the owners effectively mounted the cameras, and pay for them, but Flock got to use the footage. This particular connection was broken earlier this year, true (I'd forgotten/missed that) BUT the point still stands about what all those cameras can see - and do we really expect (people like) Flock *not* to try again? Perhaps without bothering to offer any "advantages" to the home owners and therefore not seeing any point in, well, letting anyone know? Or just Ring's parent company seeing this as an opportunity...
> used for enhancing the glasses-wearer's experience (Augmented Reality), and the data was absolutely NOT used for any other purpose (personal identification
Personal identification can be "enhancing the glasses-wearer's experience" - it can even be an assistive technology: people can be (nearly) incapable of remembering names, and basic information, of people, even ones they've met multiple times before. Having the glasses identify the individuals and putting up a name tag would allow social interaction and a normal life.
Heck, I've wanted that ability for all the decades since I learnt spectacles mounted displays were feasible, even if the cameras were rather naff back then and processing would be slow.
Flock cams are not owned by "the authorities".
They are installed and owned by the householders and Flock sell those same householders the intrusive services.
Yes, the authorities do also try to take advantage of Flock cameras, and all the others. Sometimes. And un the stories that you read about and get rattled by.
But if you don't like all those doorbell cameras watching all the public streets, bus stops etc, without ever paying attention to (in the UK and EU) laws about signage warning you are on CCTV, then have a go first at the people who are buying and installing them like that: they are not "in authority". If they installed the cameras properly, only looking at their property, then "the authorities" wouldn't find them so useful to misappropriate.
For decades now, there have been headmounted (i.e. on spectacle frames, because those are easy, convenient, self-aligning...) sensors for assistive tech. They started out, frankly, huge, nigh on ungainly: a triangle of ultrasonic transducers right in the forehead. Not widely used, but have been spotted out in the wild.
Nowadays, instead of ultrasonics (which have improved but are still quite large, for physics reasons) we can mount time-of-flight laser sensors; even hobbyist retailers will sell you a nice little 8x8 VL53L5CX sensor. Instead of a triplet of whacking great gold coloured transducers all the onlooker will see is a tiny hole in the spec frames and, if the light catches it just right, a glint of reflection. Discrete and comfortable aid for navigating a hazardous world.
Guess what else is just a tiny hole with the odd glint?
Even the entirely rational, well balanced and, dare I say it, suave and sophisticated El Reg commenters have been known to talk about "punch first, ask questions later" when it comes to "glassholes".
Hey, no problem, I guess we'll just make all the assistive tech bright orange, with a bouncy flag on a spring to let everyone know they are safe to have around. And that'll help my new-old-stock Deely Bopper sales.
Horses' eyes are large - MUCH larger than these cameras - and are "deliberately" placed to separate their fields of view, resulting in only a tiny overlap at a distance.
Earbuds can - do - have portions that stand proud of the ears and are visible when you look at the wearer from the front. The camera lens can be mounted, forwards-facing, in those locations. Yes, the region of overlap will be forced forwards from the wearer's own eyesight (get nose-to-nose and the cameras will only see the interloper's own earbuds recording the encounter). But from any position where YOU can see both of a person's ears, THEY can get a stereo view of you from their earborne cameras.
The bans are being applied to *all* glasses with cameras. The specific model named here to are only the latest, cheapest, ones from Meta, which reduce cost by not providing the display (and expecting you to feed the images into their AI, e.g. to read signage out loud to you, after translation, - which can be an accessibility feature).
Different models, from Meta and other manufacturers, do include displays, of varying sizes, resolutions and number.
> The Google Glasses display is so resolution-limited that I dismissed it as a possible wearable-computer display.
You can do a lot with low resolution displays, even those that only cover a small portion of the visual field; even as a wearable-computer display (after all, we got a lot done with low-res displays in the Olden Days). Especially if you add even a very simple beep alert. If you give up trying to use the displays for reading modern websites, low-res can outline a step down on the floor or put up a warning icon (emoji these days). Low visual field coverage, with the beep to indicate you should start, can be "enhanced" by your moving you head to point the thing all around until the dangers have been indicated.