Posts by that one in the corner 4459 publicly visible posts • joined 9 Nov 2021
We are told in the revived Dr Who that the da-da-da-dum is the sound of a Time Lord's heartbeat.
It wasn't until let in TOS that we learnt Time Lord's have two hearts (Pertwee, IIRC).
So clearly Mr Grainger was the right sort of person to write the theme on day one. Hmm, maybe getting clonked on the bonce was him trying to get into the wrong piano: his was larger on the inside.
> getting bogged down with exabytes of data is the exact opposite of being fluid and responsive.
Not to mention that, to be useful in training, you need to tag the data to at least indicate if that was a good or bad result: comparable to the tagging of training images to say "this photo contains a cat", "this is a dog" - oh, and that cats and dogs are not the same thing. Tagging takes a *lot* of effort by humans[1], doing it for all of those exabytes...
Hmm, maybe they could insist that the pilots/sappers/able-seamen type up notes as the mission progresses? "Dear Diary...".
[1] which is why, before LLMs[2] were the only thing being talked about, articles were telling us that nets were trained on such and such curated collections of images; even the bulk scraping of web images relied on them being minimally tagged as "my photos of trains" or "actor X at the Oscars".
[2] LLMs get away with more bulk scraping of absolutely everything because, to start with, you assume that anything published is tagged as a "good" sequence of characters[3][4] and fingers crossed the web page content tags indicates "en" versus "es".
[3] yes, yes, the "real" ones tokenise with bigger units than just characters etc etc; hand wavey simplifications, ok?
[4] "good" in someone's mind, at least (sideways glance at Reddit posts, eyebrows raised)[5]
[5] this has gone off track, these exabytes of sensor data aren't going to be fed into LLMs, they'll go into a neural net with appropriate pre- and post- processing... Oh heck, they'll just dump the lot into ChatGPT, won't they!
This is definitely a Good Thing (both for direct patient application and research, especially if - when? - cost reductions occur as the tech is moved from research to daily application).
But, it was a bit of a surprise the way the article suggests that they haven't included target tracking from the very beginning.
Creating an algorithm to calculate the target positions on the complex shape of a skull is obviously required, and good job solving it. However, once you have the 3D description, including things like the angles required for the effector to be - effective, isn't passing that through a tracking and retargeting layer a pretty well solved problem these days? If the piece had said that they were refining such a layer would be one thing, but the implication is that they have not started it - and are just thinking about one of the possible methods, visual tracking.
Which makes me wonder if their roboticists could have been chatting with a few other groups besides the precision mechanics guys.
> and if what I was looking for didn't exist, it would say as much rather than giving me what it thinks I might have wanted instead.
And if what I was looking for DOES exist, it would say as much (in the first results) rather than giving me what it thinks I might have wanted instead (for page after page before reluctantly indicating the right answer).
> Any responsible deployment of AI technologies requires rigorous assessments to ensure the data and outputs cannot be misused.
Misused, as in taking the LLMs results at face value!
Interesting new problem is added to the mix: how do you verify that the data Copilot presents you is the *complete* set of "relevant* data? With no little additions?
Whilst we are here, how stable are the results of your Copilot enquiry? Is the work repeatable or will an innocuous change in the way the query is worded going to cause a different dataset to be returned?
> Yeah, I'd love to have it on my phone - set to answer anything coming in from a blacklist.
Aha! What you want is a CPU with a built-in NPU so that your PBX software can run a Daisy[1] model for you.
And to think, we've been mocking Intel and MS for pushing this stuff on us, but they knew there was a use case we could all get behind.
[1] ok, maybe just a Daisy-lite model: it only has the one grandchild and can witter on about about knitting but just never got the hang of crochet.
> tests of the processor rate it as inferior to Intel and AMD desktop chips released three to five years ago.
Gee, a laptop's processor gives inferior grunt compared to even a slightly older desktop's processor. Who'da thunk?
I note that we aren't provided with any power requirements for the new device - so is it competing thermally with the desktop chips or should it be measured against, say, the Intel N100[1]? After all, you point out they came from old Via designs and those were popular due to their low power requirements.
As for the chuckle over putting it in a flashy case - why does *any* CPU get put into a flashy case? Solely so that you can put your bling in front of the sort of people who are going to be impressed by the bling. After all, who cares if the latest, greatest, CPU is cased in carbon fibre or in dull grey (or nostalgic beige) if you are actually interested in using it and not in just being a Flash Harry?
So there's your answer: this is for someone who wants to look the part. Just like anyone who buys premium-looking (and premium-priced) goods. Not really a great conundrum after all.
"But, but the insides don't live up to the packaging!". Well, ignoring how often that happens anyway, with all sorts of goods, don't forget the value of boasting that you are patriotically buying local: if that helps you seal the deal (or avoid being outcast), job done. Proudly Made in the Good Old Provinces of China.
[1] 4 cores, 4 threads; bought one in October, totally outmatched by my 2020 Threadripper box, what a surprise.
> flaw in ... Visual Studio that could be exploited ... "by loading a specially crafted file into a vulnerable desktop app,"
Is it just me, or is that suggesting that you could be caught just by trying to use VS to look inside a file you are (rightly) suspicious of?
If I've got a file that doesn't seem kosher (is that really a zip or something else?) I tend to look at it in my favourite "programmer's editor", expecting that to just show the contents and do nowt else exciting (well, show in various formats, like hex or highlighting keywords). This appears to suggest that, if I decided to favour VS and use it like that...
> And why would that code already be executed when just selecting the file?
Because selecting a file in Windows Explorer triggers far more activity than you'd expect at first glance. Heck, given the amount of stuff you can get Explorer to display in the details list, which involve examining the file contents, I'm surprised there isn't[1] an exploitable bug that fires when you just list a directory containing a malformed file. Or in a tooltip, do just pointing at the file for too long...
Too early in the day to try and find more details on this specific CVE, but if the leak is NTLM-related then wild guess it is triggering network traffic that can be be manipulated in some way (hmm, it would be nice to have links to PoC writeups to make this clearer to understand - but then again, perhaps not until everyone has had time to update...).
[1] ok, go on, tell me the CVEs that I've missed/forgotten...
> ... watch computer-science lectures from the world's best universities ...
And very good they are; I have a note on my to-do list to hunt down the "Structure and Interpretation..." lectures as a refresher (some things just don't go out of date).
> Online tutorials...
More of a mixed bunch than those lectures, but there are good ones; just don't sick to just the one per subject, unless all you really need is a worked example to do a quick job today.
> But we should probably stick to waiting until they're 16 ...
Now worried that you are against younglings watching CompSci lectures, running a decent OS etc.
Oh, no, wait - that last paragraph, you aren't trying to be sarcastic, are you? And in a direct response to my comment? But, but, that would surely mean...
Oh my goodness, you are! You are trying to equate Stackoverflow with "lectures from the world's best universities"!
Bwaa-ha-ha! Oh, glorious.
Hmm, wonder if I can find an online lecture from MIT that discusses the concept of "Signal to noise ratio" for you.
> block them from stackoverflow
Absolutely.
They need to be given the chance to learn for themselves what it takes to solve problems and what it looks like when they are speaking from a position of understanding how their answer works.
Then they can be introduced to the world of chicken-waving (and willy-waving) voodoo copy'n'paste points-scoring "solutions" that permeates Stackoverflow when they have a fighting chance of being able to spot the gems of rationality that still lurk there.
True, there are worse forums of scum and villainy out there (/Linux may be reasonably safe but that is next door to so much other Reddit...). It may be sensible to let them into Stackoverflow first, under supervision to start, and gradually phase in exposure to over sites.
Can we see a case arising where the defence is that the PO *did* just rely on Horizon and Patterson takes the stand to formally declare (whether as a witness for the defence or as one for the prosecution that gets neatly cross-examined) that Horizon is an untrustworthy witness?
Which would basically translate to a boss admitting under oath that his product is rubbish.
Would we enter a Golden Age of Product Honesty[1]?
[1] No, we'd just see even more get outs in licence agreements and weasel clauses in contracts to the effect that they won't even back their product's ability to hold a stationery cupboard's door if you print the invoice, fold it up tight and stuff it in by the jamb.
Only if there is a genuine risk that the two may be confused in the public mind.
For example, if MS decide to use another upbeat version of a well-known tune to advertise their wares and it becomes inextricably linked in the public mind whenever they see "WI" and hear "Jammin' Jerusalem".
The one with the boutonniere made from empty Beech Nut husk clusters, thank you.
After learning from two successive sentences in TFA that "Apple Intelligence" can use the initialism "AI" (always worth repeating the major point, in case we miss it), there are not as many ways to extend "AI"[1] whereas "WI" has so many obvious applications...
Windows Intelligence Neural Knowledgebase: Learning Edition (complete with pins).
Windows Intelligence Turing Complete Horology (the long awaited replacement for the Win'7 clock widget).
The sales and marketing application, Windows Intelligence Driven Data Lake Extractor.
And, of course, the new online assistant, Digby, from Windows Intelligence Generative Adversarial Networks.
The list goes on (and probably will).
[1] Apple Intelligence Modified Language Engine Support System? Ok, there are quite a few, but not as many.
from yet another old destroyer of /
How about, as a safer alternative, starting out with:
cd $QATOOLS
rm -rf bin etc incl var
(and, if you haven't invoked pwd in your prompt, then you can add one just after the cd)
A lot less typing, no daft putting a / in front of bin etc (etc) and risking killing off the key OS files. If you didn't spot that the cd "failed" then, depending upon your shell, you either just went to your logged in user's home directory or simply stayed put (which isn't, of course, in the root directory - please!). Both options hopefully far less damaging to the system.
Of course, all the other tricks can also be employed (see above, doing ls, using a script...)
> What does spending all that time in front of the computer actually get me?
You? Nothing much. You're settled and well situated with plenty of useful work to keep you occupied.
Someone 1/2 your age or younger, a bit fleeter on their feet? A few quick gigs flogging "AI to Suit Your Business" to any longer local companies with their own websites that can be convinced they require a chatbot to Join The Revolution. Being fleet of foot is needed so young laddo can skip town and find some more mugs elsewhere before the annual returns are in and the "clients" find out exactly how much those LLMs have helped the business.
An unrepentant Moist Von Lipwig would have had a field day with tuned LLMs; it is barely even dishonest, not like waiting for the Cabbage Bank share certificates to become dry to the touch.
Not with a request or a demand, neither appeal nor invitation;
No entreaty, we did not beg, beseech or implore, no levy on your time;
It came not with a proposition, no petition or solicitation;
Not an order, nor a command, no imposition upon your person.
It should have been a stipulation, our singular ultimatum, to exact your cooperation, please accept our supplication.
> I can't think of a reason to open 60 emails at once.
You have just been added to the CC for an email discussion that has been going on for a while.
Every single sodding reply has been top posted[1]. And each one had some techie detail that you are trying to cross reference against the next (too much just to assume you'll keep all those part numbers and contract IDs straightin your head without constantly looking back at them).
As the last person in the chain, you now have 3 working days left to develop The Solution that sales have been keeping to themselves for the last 4 months. GO!
To read everything in a sane fashion, you open that email, scroll to the bottom; place this at the top of your portrait screen (this is not your first rodeo). Open it again, put new copy under first and scroll to reply #1. Open it again, put new copy under previous and scroll to reply #2. Open it again, put new copy under first and scroll to reply #3; note that this one has a digit different in the part number - a mistake? Open it again, put new copy under previous and scroll to reply #4.
Repeat. And again.
After an hour of this, you have almost finished working out what the bleep is going on (yes, that part number change was an error, but it was only spotted in month 2, but the contract IDs really did change). Wipe sweat from brow. Acknowledge the presence of the other Devs, now gathered around your chair, open mouthed at the sheer audacity of trying to decode in one sitting an "all the requirements are in here" email.
Prepare to open message again and scroll to reply #103 (you closed the week's worth of arguing about whether fire should be fitted nasally) when BANG (or rather, phut) and Outlook crashes. In a Classical style, with a Geek Chorus of moans from all the occupants of your cubicle.
[1] one of the greatest sins that Microsoft has committed, convincing people that top posting is the way to go.
> Control access for systems of any significance should be completely isolated from access by non-specialist workers
Good point in general, but in this case the keyboard was clearly expected to be required (hence the shield instead if, say, locking it in a cupboard) and it was inside the existing control room, full of tasty knobs and buttons, so one would presume that everyone on the shift was already relevantly qualified and specialised personnel.
Although perhaps, for the sake of the new system, still a bit too much in the hefty gas rigger mindset and not willing to give up dropping me sarnie box on the table like I done for the last five years before this interloper arrived. A bit more "sensitivity towards the fragile new equipment" training was also in order.
> we found counterintuitively with Open Compute that by publishing and sharing the architectures and designs that we had for our compute, the industry standardized around it a bit more
Sorry, Zuck thinks that it would be more intuitive for industry to standardize on something whose architecture & designs are a complete black box to them?
Okay, it works for Microsoft (they'll change the file format for Word and then not bother to properly document that) but for most things we'd like to know a bit about what we're buying, wouldn't we?
>> Thus far, AI haven't been able to explain their reasoning.
AI techniques such as Expert Systems have "explaining their reasoning" as a core part of their design.
The entire purpose of Planning systems is to create the explanation of what they are going to make the systems they control do.
Neural Nets that are processing images have been able to be examined, to illustrate (literally!) what their layers are doing, the features they are triggering on, and this has been used to explain how their outputs were reached.
It is has been the lack of any explanatory abilities that have been one of the issues with LLMs and how much we should rely on their answers.
> Why are we still using terrestrial instruments for astronomy?
> Learn how to put your instrument on a satellite and launch it.
Where do you start replying to someone who has clearly never bothered to spend even ten minutes to learn what astronomical observation is all about?
Who can not even distinguish between "an instrument" and a telescope that you can attach the instrument to?
Or spend another ten minutes learning (or just thinking) about the costs (in all areas) of using space-based instruments to deal with the lack of available and upgradable resources (power, maintenance, in-situ compute, planned upgrades, unplanned upgrades when a post-grad makes a stunning discovery).
Or just find out the difference between the number of space-borne 'scopes and instruments compared to the number of ground-based (oops, you just counted the Big Boys in Hawaii and forgot all the observatories in Universities, schools, local astronomy groups, back gardens - and the ones not in observatories, just taken out when possible...).
COBOL?
Nah, let's go for Forth.
And do it *properly*: define layers of meta-compiler words[1] and keep going until we run out of different ways we can express brackets/parentheses/braces[2]. Then the real meat of the code can be expressed compactly, no need for multi-page functions!
[1] bonus points for defining punctuation words in layer n, using them to create layer n+1 then re-defining them to start layer n+2, because then the code always familiar and neat.
[2] hey, you can get Unicode Forth now, so proper paired speech marks, in various languages, can be used. And those funny upside down pling and question mark. Maybe we can find some pictoglyphs that sort of look like left/right pairs to an English speaker. Ooh, ooh - smiley and frowny face emoticons! We can use a school bus to be a bag of values that is driven into wall to do a merge sort...
> C was a lazy choice
For Linux, C was a very sensible and pragmatic choice.
Now, if you (or anyone else) had been able to tell Linus how large Linux was going to grow, *and* given him the money to buy the compiler and kit to let him start with the better choice for his personal project that wasn't going to be a big and professional OS like Minix...
> However, I have many doubts that Rust is the language that will be the one to take the place of C... What it'll take is a major paradigm shift
All this argument over Rust is the sound of people trying to make a paradigm shift without disengaging the clutch[1]
[1] or even having built a finished clutch in the first place, to overstretch the analogy and strain the joke.
> The thing that always gets me about computer scientists ... the first thing they do is invent a new language to try and solve it in ... Rust
You are aware that Grayson Hoare created Rust as a side-project whilst working as a Dev at Mozilla?
He wasn't an academic Computer Scientist.
CompSci certainly creates plenty of new languages - it is a good way to make your experiments[1] concrete and demonstrable, especially when publishing (and creating a thesis!). So, yes, of course CS invents new languages (or modifies existing ones).
But very, very few of those are ever even intended to "go mainstream", certainly not as-is. The *concepts* are hopefully going to end up somewhere practical, but like the bulk of science, what Joe Public ends up using day to day looks little like the original publication.
On the other hand, the languages you actually *see* as a Dev grunt[2] tend to be - rather more pragmatic. PHP, Perl, Python, JavaScript - yes, even C and most definitely Rust.
Curiously, you can tell - by looking at the language specification. Or the lack thereof.
To get your CS thesis best marks, you have a specification, and you use comparisons of this to point out where your nifty ideas are new and different.
To get your personal side-project out there, you fling together all the example programs and the compiler/interpreter executable, then try to keep up with the various totally random demands[3] that (re)shape your baby. After a while, if you (and everyone else) are lucky, there is enough interest that (other) people start work on a proper language specification, for pragmatic reasons.
[1] as in, the actual "science" bit.
[2] except for the lucky few who actually get paid a salary, not just a sequence of grants, to work on production code in an "obscure" language, such as ML.
[3] like JavaScript changing its outer appearance to look more Java/C'ish whilst underneath... ok, JS is a rather pathological example...
> Some users are reporting the issue as worse than has been officially acknowledged and has far greater impact than first believed.
Have you got any solid URLs to share?
I've seen chatter, incuding some doomsayers, but nothing reliably techie (comparable to the original report from Ian Lesnet).
> It is not raspberry pi's first rodeo when it comes to getting things wrong
Sadly, definitely seen "reports" and "discussions" that take that premise and then catastrophise, which makes it harder to find out the realistic state of play.
But, so far, my sole RP2350 has been happily working as a direct replacement for an RP2040 when plugged into a happy project, except that it provides more memory resources and some more general oomph. So if you *are* chucking away a cart load of dev boards then I'll still take 'em of your hands,
> or connected to the serial port or USB port
> you connect to via ethernet
Ah, so the *important* thing about the modem is whether you are connected via a direct 1:1 (serial) cable, by a 1:n network (USB) cable or by an n:n network (Ethernet) cable. Gotcha. And there I was thinking that the important thing was signal modulation.
I wonder if there is anything else that we can therefore declare "obsolete" because they are now (most often) on the n:n network - printers, perhaps?
> that's not what I'm talking about.
It is always good to be precise or you are just going to get people arguing with you.
> Modems are obsolete now
Alright, Miss Smug "I only use fibre"[1]. Some of us are still only on ADSL you know. That box in the corner may connect to the LAN and not just into one PC, but it is still a modem.
[1] although conversion from electrical to optical modulation still counts in my book; renaming it is marketing, not tech. Ooh, and there is radio modulation as well.
But DNA did also bring us a vision of hope:
"Bring out your dishwashers! Bring out your digital watches with the special snooze alarms! Bring out your TV Chess games! Bring out your Auto-gardener’s, Technoteachers, Love-O-Matics! Bring out your friendly household robots! Shove ‘em on the cart!"
Although I have my doubts that we will ever reach the level of artistic achievement that allows the nullification of Gravity; it would save us so much effort, even if all the lauch vehicles did have to look like Nutrimatic cups.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
