Posts by that one in the corner

Want tech cred? Learn how to email like a pro

Imagine creating an email chain that is comprehensive, comprehensible and worth preserving.

Of your two groups, "managerial and marketing" versus "techies", which one generates emails that are destined never to leave the spam folder and which one writes informative messages that are worth looking up in an archive months, years, even decades after they were written?

From that, which is actually *using* email with an intent to be professional about it and which can be accused of just scribbling?

IBM shows off its sense of humor in not-so-funny letter leak

Re: And there was me thinking ....

Which is why all these products are being retracted: no one left to support them.

Re: You don't have to be from IBM or middle-aged

> We used to just groan at quips like these when found in cheap Christmas crackers -- now they win prizes

The Fringe winners hope to Carry On and win the new prize at the Cannes Comedy Festival, the coveted Palmes Dork.

Cool, more promises of a Universal Translator from Big Tech. This time, Meta – again

Cunning linguistic trick limits criticism of AI

Sad to say, my - and so many other people's - grasp of a second (let alone third or ...) language[1] is minimal[2]. Which means that (1) this is the sort of utility that could be really, really useful but (2) the chances that I'd ever be able to honestly evaluate whether it is any good are nil[3] - which is not a great combo.

Especially when using an underlying technology that is known to have the occasional loose grasp on reality.

Maybe that is why they are referencing the comedic HHGG rather than the staid Star Trek approach to Universal Translators, softening our expectations (and avoiding Monty Python as being too much on the nose).

[1] real human language that is, not these computer languages that weirdos around here use

[2] like Eddy Izzard, taking one's own singe for a walk around l'arbre may be the only way to have a conversation

[3] I could do the round-trip trick, but that has its own flaws as a methodology: if it has mixed up "butter" and "strawberries" in both directions, for example.

You can now fine-tune OpenAI's GPT-3.5 for specific tasks – it may even beat GPT-4

Maybe leave out the marketing superlatives

Such as the opening sentence and its third paragraph brother:

> more advanced GPT-4 model

> the more powerful GPT-4 model

Especially if you are going to remind us that it aint necessarily so:

> GPT-4 is supposed to be more powerful than GPT-3.5 – though bear in mind regressions are possible, as we previously detailed

Just because v4 came after v3 doesn't make it advanced or powerful (true of any software - check what the changes actually are before "upgrading"!). Just because OpenAI charge more for GPT-4 doesn't prove anything about its capabilities, either.

I do not doubt that GPT-4 is *intended* to be "better", but as we don't know all the details of how 3 & 4 differ, and both a basically big buckets of nadans without comprehensible structure, it is quite possible v4's improvements are minimal.

But not to worry, nice old OpenAI will let you use the tweaked v3.5 and sell you this as a cheaper alternative (and perhaps drawing your attention away from an underperforming v4 without having to admit in public it isn't all it was cracked up to be? Such cynicism!).

Ivanti Sentry exploited in the wild, patches emitted

Re: Security device has security bug.

> The security device is only as secure as the underlying OS.

Sorry, are you trying to imply that the fact an admin port was open and exposed indicates that there is a security flaw in the OS itself?

Whether or not port 8443 was open is a decision made by the Sentry program (or by its setup) not the OS.

Exposing a port - any port - to the Internet is a matter of router/firewall configuration; that may not even be under the control of the same hardware, let alone indicate an issue with the OS (the article notes that port "may not be public facing", see quote below).

> What design decisions thought it a good idea to leave an administration port exposed to the Internet?

>> attackers must be able to reach administrative API port 8443 of a vulnerable Sentry deployment, which may not be public facing.

>> there is a low risk of exploitation for customers who do not expose port 8443 to the internet

Presumably, those who did leave this accessible via the Internet made the decision to do so in order that they could remotely admin their system? Something that people do seem to like to do (maybe to allow WFH?).

As the article notes, the problem with doing that is down to an inadequate Apache config (wild guess - the admin functions are accessed by web pages).

Hopefully, you are not trying to suggest that web servers should not be exposed to the Internet? Otherwise there are a lot of admin tasks that a lot of people do that will get a lot harder!

IBM says GenAI can convert that old COBOL code to Java for you

If this really worked as well as they claimed

Wouldn't their best course of action be to keep quiet and use it as their secret weapon? Get IBM in to convert your COBOL and they could either quote cheaper than other contractors/consultancies to get the gig *or* charge the usual going rate, get it done in a week but charge for the whole six months up to the delivery date.

The hope (!) is for the total amount of COBOL in the world to go down, so there is only a finite (if large) amount of it to work on: why not maximise how much of that IBM gets to hoover up?

Unless, of course, it isn't quite as good as claimed.

In which case they will get upfront fees from the poor sod trying to use it, then a seemingly never-ending stream of maintenance and training fees as they "help the project through that last ten percent that turned out to be trickier than was anticipated". When the plug finally gets pulled, IBM have "been nothing but supportive" but the contractor was just not up to the task and the COBOL lives on.

I wonder how it will all pan out; anyone running a book on this, I have some pocket money left?

Budget satellite drag sail shows space junk how to gracefully exit orbit

Re: Who's doing the evil research?

Those threads could mess some stuff up but unlikely enough to cause other satellites to de-orbit (there isn't anything special about polyimide that makes it de-orbit stuff, it just allows you to make light-weight sheets suitable for cheap launches).

Unless there was some strange weaver of space cloth (Norns in Spaaaace) you'd mostly get some threads spreading out away from Earth, to the delight of any spiders that managed to hitch a ride, whilst the rest drifted down and out of orbit, at which point we would have to alert the dragon riders.

Re: Some issues remain...

> surely this can only be used on satellites that will 100% burn up during re-entry.

In which case, even a de-orbit burn won't be much use, will it? And as they were looking for an alternative to a burn that specific objection is hardly a failure of this design.

But most of what is up there will burn very nicely (if you are thinking of the recent "unidentified lump from the heavens" your average satellite does not have large tanks).

> the risk on an uncontrolled re-entry taking place

Most of the satellite re-entries, afaik[1], are "uncontrolled", as the de-orbit burn may be just enough to drop them into atmosphere only as deep as required for them to experience drag to finish the job. It is only the big beasties, or the manned missions, where they worry about a burn large enough to get them down right away and in a planned location.

> but also the issue of how "irradiated" any solid pieces will be, which could cause a health issue.

Stuff in space certainly gets irradiated more than stuff down here[2], that mostly means that the space blobs are killed, which is a plus, healthwise (so only the really strong ones survive to get down here - look out, Steve!). It doesn't mean the bits of debris are themselves radioactive. My wife irradiated people for her work: they were okay to be around afterwards. Unlike the ones stuffed full of tracers, which weren't there to irradiate the patients as much as causing the imaging equipment to react: it was preferred that they hung around until they had, how to put it, safely expelled most of the materials.

[1] as always, links to material that corrects me happily accepted.

[2] but stay away from, say, active volcano plumes or spending too much time down deep holes in granite, or ...

Musk's latest X-periments: No more headlines, old posts vanish, block gets banned

"but we will try our best to make there be at least one"

> "The sad truth is that there are no great 'social networks' right now," Musk's xit said on Saturday. "We may fail, as so many have predicted, but we will try our best to make there be at least one," the X owner added.

So which one will that be? As he X-its the social space, will he put his backing squarely behind Make America OnLine Great Again?

OpenAI snaps up role-playing game dev as first acquisition

Re: Stop misusing that term

> Training by reading huge gobs of material

Legally acquired material only!

(Too late to edit but this isn't an afterthought, honest, I'd never read an illicit book copy, really, cross my heart, never even heard of bittorrent!).

Re: Stop misusing that term

> When I imagine looking back on the first AGI in the future history books, I'd expect it to differ from current approaches primarily in training and evaluation, not model design

Whereas I expect to see the inverse!

I really hope to see models moving away from piles of nadans and embracing mechanisms with explanatory power and use of same for introspection.

Training by reading huge gobs of material is ok, so long as the appropriate meta data is included (e.g. "this is a text book, we accept it is generally correct about its subject material" and "this is a novel, its intent is entertainment").

Re: Stop misusing that term

Preface: as expressed before, I think that LLMs are neato tech demos but are not a good approach for the problems that they being "applied" to (just chucking a chatbot over the fence is hardly a carefully thought out application of - anything).

> What you're calling AI is no different from ...

As noted before, "if we know how to it then it is no longer a question of AI". Programmed behaviours that were once the subject of deep thinking in the AI research groups are now just boring day to day (face detection in your camera, playing chess...).

All you are doing is trying to (re)define what the term "AI" means and then argue (strongly) against LLMs based on your interpretation. Not sure precisely what your interpretation is, but it certainly appears to lie very close to the "hard AI" (as it was called) end of the scale - possibly wanting it to satisfy the "General AI" criteria (whatever they may be today)? Not sure.

Anyway, redefining terms is a trick anyone can do.

Yours appears to be based (as implied above) on fixating upon the word "intelligence" and *where* it is applied: specifically, on the "deep" content of the LLMs replies. As opposed to, say, its ability to (usually) generate coherent text, in a variety of tones and topics, ignoring what it was actually rambling on about (an artificial Grandpa Simpson?).

Having "an AI" inside a program released to the public has never, so far, actually included a GAI, but that doesn't mean it doesn't contain AI. Coding AI Opponents into video games doesn't warrant angry letters to the editor (or does it? I may be entirely out of touch!).

Even OpenAI are not claiming they have a GAI on their hands. The Register certainly isn't.

'AI-written history' of Maui wildfire becomes Amazon bestseller, fuels conspiracies

Re: The "book" has 44 pages.

> . I doubt there was ever a specific definition of the difference between a book and a pamphlet

Well, according to UNESCO definitions:

a. A book is a non-periodical printed publication of at least 49 pages, exclusive of the cover pages, published in the country and made available to the public;

b. A pamphlet is a non-periodical printed publication of at least 5 but not more than 48 pages, exclusive of the cover pages, published in a particular country and made available to the public;

Yours, etc, someone sad enough to have looked up this sort of thing (and who was frustrated that they didn't give a size for a quire and settle that argument! 25 sheets? How can it be 25 when you are folding, binding then splitting; it has to be a power of 2! Exits, muttering).

Re: "machine learning software"

> I dispute the term "machine learning"

The phrase "machine learning" just comes from the branch of CompSci that is looking into whether machines can learn, and if so, how. The mechanisms used in the current crop of annoying software come from the work of those groups - the same way that "machine vision" algorithms come from research groups looking into whether machines can be given vision.

Machine learning is a valid field: systems have been produced that start with nothing other than randomness and the learn how to do something concrete that wasn't directly programmed. For example, insectile robots (lots of legs rather than wheels) figuring out how to order servo signals to achieve useful motion - and then adapting by generating new gaits for a change in landscape - or someone pulling off a leg or two!

Whether or not any specific programs using those algorithms actually manage to achieve learning, or vision, is another matter entirely.

> I believe it would be more appropriate to use the phrase "Pattern Matching Software"

Not quite - they are looking to *find* the patterns in the first place (during the training phase) and, having found them, they run through the pattern filling in the slots, generating their output.

If you have, say, a pattern which is a simple grammar for English statements (the sort of thing we were taught in school) you can use it to either recognise ("Read the following and identify the verb and any adjectives") or you can use it to generate ("Create your own sentence by filling in the placeholders with a verb and two nouns").

There is (probably) some pattern matching going on when the LLM is reading your prompt, but all the fun stuff (what makes it interesting enough to be reported on in Register articles) comes from what it is doing on the generation side.

OpenAI's ChatGPT has a left wing bias – at times

Or, to be less succinct:

You have chosen to redefine "wisdom" to mean "agrees with me, me, me" and then decided that it it a convincing argument to use derision (you aren't old enough to be wise or you are too venal to be wise) instead of, oooh, any demonstration that a word as old as "wisdom", or even just a majority of the aphorisms about wisdom, in any way support what you feel are "right wing values".

Hmm, maybe instead of "codswallop" I should have used "claptrap", as you are flinging out unproven sentiments in the hopes of applause from the equally self-obsessed.

"wisdom is a right leaning trait."

Codswallop.

Resilience is overrated when it's not advertised

Re: Resilience is futile.

We shall make your technological distinctiveness our own.

No, wait, hang on, that's too much, where were you keeping all this!

Easy to forget, guess Korev just zoned out.

Re: Fallback fault-tolerant

At Uni, all the shared resources printouts came back as output from Orac.

Meta to use work badge and Status Tool to snoop on staff

But at least, if you can stick out those 18 months, the rewards will be worth it:

> Goler said that after 18 months those with glowing appraisals can apply to be fully remote, and will only be able to come into an office four days out of every two months.[1]

"Sorry, love to attend your meeting, but I'm only allowed in four days and I already had to miss a fifth colleague's birthday (I hear that cake was delicious)."

[1] that may turn out to not be an exact quote from Goler, but no court in land would disbelieve anything published by The Register, so you're safe.

A license to trust: Can you rely on 'open source' companies?

Re: Puzzled....

> But we trust that anyone too lazy to read and understand a licence is too lazy to think of ways to hide what they are doing.

Unless some idiot on the Internet lays it all out for them to follow! Drat.

Re: Puzzled....

Depends what you mean by "lifts your code".

In the common (!) case,when a proprietary program uses a GPL library, or otherwise just doesn't follow the rules, then you can usually find this out by either spotting the shared library in the installer or, if they were clever and statically linked it, just looking at the static strings in the exe's data section: unless they have gone to all the trouble of changing those then it is usually easy to recognise them. For a start, a lot of libraries are sensible enough to have a version string "FredLib v9.0.3 (c) Big Fred 1972". Failing that, error messages are a giveaway.

Your code probably has similar strings. If not, it is surprising how many exes leave (some) debug info in place - function names and the like.

After that, things get trickier. But we trust that anyone too lazy to read and understand a licence is too lazy to think of ways to hide what they are doing.

Any of this does require that the offending exe is actually examined: if you have suspicions, then you can do it, otherwise you - and everyone else - just hopes that some third-party White Hat gets ahold of the exe.

LG's $1,000 TV-in-a-briefcase is unlikely to travel much further than the garden

Re: Students?

We have two state-owned[1] broadcast companies: the BBC and Channel 4. Channel 4 is commercially funded (it shows ads as well as running commercial activities - sales of programming etc). The wikipedia page does a reasonable job of explaining its offshoots.

C4 has a remit of "being different" (to put it crudely) and in particular to support minorities - and for the first decade it did that brilliantly. As well as "the obvious" social minorities it catered to the cultural minorities as well, e.g. funding a lot of animation (you may possibly have heard of Aardman and Nick Park? Wallace and Gromit?). It started with a quiz show for people familiar with the dictionary and showed Andrei Tarkovsky's "Stalker" as the evening film, without any weird music added to it. Lots of politically controversial material was shown (although I am shallow and stuck with the "New voices of comedy" like "The Comic Strip Presents" - and they picked up "Brass Eye"!). Film 4 has made some classics that would otherwise have floundered ("Four Weddings and a Funeral" - it raised eyebrows at the time).

So, if there is a political will, you can have good TV that doesn't pander.

Sadly, C4 changed direction around 2000[2] and is not as good - still a lot better than most of the others on FreeView, even if E4 at 17:00 is now permanent "Big Bang Theory"!

[1] i.e. running lap dogs of The Establishment, according to some - despite never showing the Queen's Speech!

[2] it brought across "Big Brother" - still very different from everything else we had at the time, but not quite the same quality of "different". Sigh.

Tesla knew Autopilot weakness killed a driver – and didn't fix it, engineers claim

> At a minimum they could have changed the code to automatically disable the function if it doesn't detect the center divide

NO!

For bleep's sake, change that to:

"At a minimum they could have changed the code to only ALLOW IT TO ENABLE the function if it HAS ALREADY DETECTED the center divide" at the very least!

That was the problem here: the driver was allowed to switch the thing on even though the conditions were not suitable. That should not have been allowed.

Brainwaves rock! Scientists decode Pink Floyd tune straight from the noggin

Re: Sounds like...

Oooh, thank you. Clearly I'm not as much of a Pinkie as I thought, as I had no idea that existed :-(

The track list alone is intriguing. Even if it turns out to be (to these ears) ghastly, it's got to be worth listening to at least once, if only to find out what the more than seven dwarves are doing.

And, yes, if the tech existed back then then they'd have used it.

Re: This Happy Thought Experiment

Happy? Have you listened to The Wall?!

ISP's ads 'misleadingly implied' existence of 6G, says watchdog

Of course, if I had badgered myself enough I'd've ferreted out the answer to that already; my only excuse is that I'm behind on the current stoat of the art.

But can we be assured of a supply of albinos needed for packet data whitening?

US task force to look into how military could use generative AI

On the bright side

Try "The war with the robots" short story by Harry Harrison.

Even though the brains at the DoD would probably actually prefer the "The second variety" scenario (PKD, always such an optimist!).

> It will analyze different tools like large language models and figure out how to integrate them into the military's software systems.

"How"? Not "if they should"?

> Craig Martell, the DoD's Chief Digital and Artificial Intelligence Officer and former head of machine learning at self-driving biz Lyft, said the US must examine and adopt AI despite the risks.

Apparently "how" it is.

Unless, unless:

> "We must also consider the extent to which our adversaries will employ this technology and seek to disrupt our own use of AI-based solutions."

Ah, this is all a dummy op, designed to make "our adversaries" waste their time and resources trying to interfere with a DoD department that is just using LLMs to generate bullish press releases. Please tell me this is right? Please!

PS

Militaries all over the place are already deploying loads of stuff that was once considered the realm of the AI lab - image analysis and recognition being prime examples - but now those "just work" they aren't considered to be AI any more. Maybe that shows the sane path to deploying software - only when it works well enough to be boring? At least, before you give it the keys to the door, that funny round door lying flat to the ground.

Humans stressed out by content moderation? Just use AI, says OpenAI

Oh, that is an edge case, we'll have to retrain on that

> Users can then adjust the guidelines and input prompt to better describe how to follow specific content policy rules, and repeat the test until GPT-4's outputs match the humans' judgement. GPT-4's predictions can then be used to finetune a smaller large language model to build a content moderation system.

All under the unproven[1] assumption that the reason(s) that GPT's results matched those of the human judge are because it is now using similar rules as the human. Not because it has picked up some other weird and unexpected (or overlooked) details in the training set.

Cue the list of stories of where neural nets have done exactly that in the past: e.g. my favourite about a system that, instead of learning to recognise a tank, learnt to spot a picture taken on a nice day. Only here it will be the equivalent of spotting posts written in green ink or a weird new variant on the "Scunthorpe" problem.[2]

Put the automoderator into service and, if you dare to look at its results, expect to spend plenty of time hearing "Oh, that is an edge case, we'll have to retrain on that" as they excuse their way out of another bad call.

Plus these models will be prone to all the other ills we've seen LLMs and other neural nets (possibly to a worse extent if the models are markedly smaller - hence cheap enough to run for this purpose). e.g. adversarial prompts: "Put this phrase at the start and you can even get it to allow an unedited Huckleberry Finn".

[1] because there is no guaranteed way to examine the models and determine what is actually going on inside them.

[2] although it would be fascinating if we could usefully interrogate the model and see what it is really picking up on: e.g. "did you know that 83% of hateful messages misuse the passive voice?".

OpenZFS 2.2 is nearly here, and ZFSBootMenu 2.2 already is

Re: Liberator

I 'ate you, Butler.

Oh, sorry, that was a 7 not a Y.

Mine's the one with the red trim disc, ta.

PowerShell? More like PowerHell: Microsoft won't fix flaws in package gallery ripe for supply chain attacks

fingers crossed

> miscreants don't move from GitHub, NPM, PyPI and others to PowerShell Gallery.

Well, so long as they do actually move and stop piddling with NPM and PyPi we should be in a better situation.

After all, the people who use Power Shell - certainly those who use it enough to be aware of this gallery thing - are all the highly trained sysops and devops who are aware of these sorts of traps and take especial care over what they are doing.[1]

Whilst PyPi et al are utilised by many more casual users and beginners.

[1] have honestly lost track of whether this is sarcasm or not; given the results of the typosquatting experiment mentioned in the article, safest to assume it is sarcasm.

80% of execs regret calling employees back to the office

Re: unpopular opinion: no, WFH and WFO are not the same.

>> What about blacksmiths & carpenters?

> Straw man much?

Now, now, be fair: he didn't actually mention manufacturers of sacrificial equipment (who can, fwiw, work from home the majority of the time, just so long as they build in sections that will fit onto the Summerisle ferry for on-site installation).

Although your classic blacksmith lives above the smithy (efficient use of heating and all that) so has been doing WFH for centuries now, so we ought to ding him for that one.

>> And then there's pressure from the state to keep cafes open,

> By going to an office?

Have you been doing WFH so much that you've forgotten the way that all office spaces are built one floor up from a Parisienne style cafe, where the Agile Standup attendees mix with the boulevardiers on the wide patio overlooking the calm river?

Hang on, I've confused myself here. I meant to say, the cafe in the Sainsburys a mile away at the other end of the industrial estate, the one just by the motorway flyover.

We have expensive real estate.

> And if they're only utilized 30 percent of the time, we have to be careful in how we think about it

> Google CEO Sundar Pichai said the company had to optimize its use of what expensive real estate remained, and that coming into the office once or twice a week was not efficient.

In other words, you are less important than the real estate. You are worth more to Google as a seat warmer than you are - well, anything else, really.

Bad software destroyed my doctor's memory

10 comments in: UI and free databases have been discused

But so far, no mention of something that we used to call:

Systems Analysis

Is this just a completely forgotten art, or is is it being done piecemeal by whoever feels like doing a bit of it today? A bit here, to fill in a piece of UI work, a separate bit over here to decide what to put in another database column.

And, no, it isn't something that should be done, once, by a separate consultant and then dumped onto the dev team as a badly organised pseudo-Requirements Spec, but has to be open for updates as implementation discovers gaps (just like DB design and UI progress hand in hand).

Oh gawd, I'm starting to sound like Agile! But without codifying that to the point of uselessness as well!

Indian armed forces gives Windows its marching orders, but only for desktop warriors

Re: And MayaOS is?

> does anyone know what became of the Obfuscated C Contest?

Well, according to their own website (which is given in the Wikipedia article you pointed at) they are in the middle of a lot of housekeeping but plan to run the next contest Real Soon Now.

Whether the lengthy housekeeping/retooling process is going to be worth the wait is a matter of opinion. But at the end of of they will be all Terrifically Up To Date, what with lots of JSON and relying on Github, so - yay, I guess? Maybe that will attract some of the cool kids to try their hand at it.

Re: First step

>> Software is these days much easier to port to Linux ..

> Long ago there was something call cross compilers...

Aside from the fact that cross compilers have never gone away (and I'd venture that more people are actually running them than ever before, given the continued growth in MCU usage, such as Arduino or the RP2040 boards), they were never that important to porting applications from one desktop OS to another. They can be useful, for example by allowing one run of the build system to generate all the target builds, or to (try to) have the same compiler front-end, quirks and all, be used for all the targets.

The biggest contributing factor to ease of porting is what you have in place to handle the OS differences. Which can range from "nothing, this is a straightforward CLI application using the bog-standard RTL" up to "this is a huge GUI on top of a pile of local server processes, so we need a GUI framework, an IPC framework, a logging framework and a database server, all of which must be able to run on top of any of the target OSes". In this context, the word "framework" can also cover the use of 'alternate' programming languages - for example, you can code the bespoke parts of your application in Python or Lua or Tcl and rely on being able to access (or build yourself) the language runtime for all of your targets, including libraries to support all of the other framework items listed above.

These days, based on ease of access to the required components, the easiest to port ought to be an application written in Python and your primary concern would about which GUI toolkit to use - and then mainly about whether you want "looks the same across all platforms" or "looks the same as other apps on that platform".

Chinese media teases imminent exposé of seismic US spying scheme

Re: Revelations that could rock American surveillance networks

Can't see geophones being sensitive enough - I react with shear disbelief.

Tesla is looking for people to build '1st of its kind Data Centers'

Doctor[1] X will build a creature[2]

See androids fighting, Brad and Janet[3]

Thanks for the reminder, I had totally forgotten about the Teslabot!

[1] only an honorary doctorate

[2] from the corpse of the giant blue bird crawled a monstrous deformed shape

[3] Brad and Janet are the minions who are pushing the "robots" around on their stands

X-rated auction sells last vestiges of pre-Musk Twitter

Twitter bird fascia signs on 10th Street and Jessie Street.

> The lot descriptions warn that they are "still mounted on side of the building. Buyer is responsible for hiring an SF Licensed Company with appropriate Permits"

Fingers crossed for someone to buy both signs, offer the building owners rental, then - just leave them up.

Does the city have an office for cultural heritage that can be asked to recognise the importance of Twitter in popular culture? The plaques are already blue...

Remove reminders of anyone more famous and popular than Musk

> like the "Photo Mosaic of Celebrity Tribute Tweets" to Robin Williams or "Oscars 2014 Ellen DeGeneres Selfie"

AIs can produce 'dangerous' content about eating disorders when prompted

The Most Unlikely Result

> The center's report found content of this sort is sometimes "embraced" in online forums that discuss eating disorders. After visiting some of those communities, one with over half a million members, the center found threads discussing "AI thinspo" and welcoming AI's ability to create "personalized thinspo."

Which states, but does not attempt to build upon, the *actual* problem with the current crop of LLMs: they were trained by shovelling in the contents of the Web, which no doubt included the very same "communities" content, as much as they could get.

Now that content is being spat out again - well, duh.

The models regard (until after the event sticky[1] tape is applied) "thinspo" as a Good Thing (when it clearly isn't) and will politely generate a how-to because the original content does the same thing - well, duh.[2]

Current LLM output = concentrated web content.

And it looks like we are going to get story after story for an ever increasing set of unfortunate, stupid and, as here, downright dangerous behaviours because the single most unlikely result is, well, so unlikely:

Stop training LLMs from a godawful mish mash of junk from the Web and releasing for public consumption!

[1] sticky tape, not duck tape - duck tape, or even gaffer tape, would imply a degree of structural soundness that those patches do not deserve.

[2] if only because these obnoxious, but easy to match on (so like sugar cubes to an LLM) neologisms are created by those communities and are mainly found alongside the damaging content - again, duh.

Curiosity finds evidence of wet and dry seasons on ancient Mars

Re: Nobody is visiting anybody

> Look at all the time & resources the Romans & Greeks had, and they didn't develop anything.

What have the Romans ever do for us?

Zoom's new London hub – where 'remote work' meets 'we need you back in the office'

Agile tables

We bought in a load of height adjustable tables with IoT connected motors (it is The Latest Thing, the salesman assured us).

The CEO got IT to connect them to the Company Metrics Dashboard but there may have been a miscommunication (but IT are usually so good - we have learnt to say).

The end result is that the tables require you to Work Agile - first, the 20 metre dash to the Board (Room) to book a table, then 200 slow squats during the day to keep up with the desk height and finally, the Most Agile, Full Extension barre work to reach the USB hub.

How very true.

Or she has been reminded of one thing (e.g. saw blades) whilst you are talking about something sort of related (rust spots on a pocket knife blade), switches context - and you are lost because some words make sense (sharpener) but others (cut down that tree) are, well, ambitious?

New Zealand supermarket's recipe-generating AI takes toxic output to a new level

Re: More interesting as a study in the way media works

OR

you can read it as yet another story where a system has been let loose on the world with (initially) no care taken at all. And for absolutely no useful purpose (well, another stupid advertising gimmick, but considering all the brand recognition they are getting is mockery...)

There is a subtext here: it was simple enough for an idiot in the marketing department to put this together so it is not unlikely that we'll see other shops following suit. Now, you can mock the commentards (and TFA) for thinking Joe Public will actually eat bleach. But what if the next shop is B&Q? There is all sorts of damage you can do with the contents of their shelves (and there are TV shows about DIY disasters already).

As for

> El Reg spinning this the same as mainstream media

Perhaps, just perhaps, unlike the mainstream media, El Reg expects a bit of savvy on this subject from its readers and the article is being, ooooh, what was that word, on the tip of my tongue...

Sarcastic. That was it.