Re: Really?
What is this, a house for ANTS?
5497 publicly visible posts • joined 9 Nov 2021
> $200/month to "monitor sleep data"?
Given that
>> EightSleep pumps out 16 GB of data a month
they clearly need to spend on their servers /s
(Although, to heck with "I wish I had the spare change to spaff $200 a month on this", I wish I could get an *uplink* capable of handling 16GB p/m! And still leave enough for El Reg to work. One day, Real Soon Now, after the cows have gone home because the last few of the neighbouring fields has been built on...)
It is a productivity booster.
Boot your PC, open the browser: Ping! "Here are your browsing results for today, now you can spend those five hours mowing the lawn; and you know I'm watching that you do[1]".
Just you wait, this browser will be installed by wives everywhere. Dishes will be done, shelves will be put up[2] (any truth to those rumours that Altman is buying B&Q?).
[1] Courtesy of the 'Net connected doorbell camera.
[2] ok, this is just a fantasy, but isn't that the theme for today?
> started with an assumption “that we would be spending time on architectural differences such as floating point drift, concurrency, intrinsics such as platform-specific operators, and performance.” ... "it turns out modern compilers and tools like sanitizers have shaken out most of the surprises"
Sorry, what?
They planned to be spending their time on dealing with minute details of instruction sets and opcode interactions? The things for which we've been building decades of maths - you know, the ONE area that can actually deal in proofs! - from formal theories of languages to planning and optimisation of completely defined[1] and constrained operations in hardware? Variances in floating point accuracy and stability, the bread and butter of published mathematicians and computer scientists; published, as in actually told everyone and provided worked examples as compiler patches to present to the examining board?
But were then, apparently, surprised to find that their time was really needed on fixing things that were rather less rigorously created in the first place, like build and release systems?
Worse (!) having discounted the fine and detailed maths described above, they hadn't planned to be tackling the bits of maths and stats that they should have been expecting to find in their own systems, such as overfitting tests. Or the issues inherent in keeping existing running interacting systems stable (hint: keep the machine room clear of butterflies).
Ok, it wasn't my best subject, but being sent out of the terminal room, away from the blinking lights, and into the lecture halls of the Department of Mathematics, Statistics and Operations Research was once seen a basic to producing a well-rounded geek.
Let alone one with the title of "engineering fellow".
And then ... they patch it all up with "AI". Run a learning model, get it to print out what it has discovered, sanity check that, then engineer the results into your improved systems, good idea. Letting it run wild on its own...
> . If you desire resilience, you have to create applications which aren't bothered when EC2 instances go away. You also need an ops team that knows how to handle the necessary traffic routing & alarms.
So you have to spend inordinate amounts on devs to do the extra clever coding and on a big and fast responding ops team to keep up with all the routine hardware failures? Certainly hope you're saving enough in the low, low cost of hiring the EC2 instances to make up for that.
And all of that required and repeated in every customer.
Wasn't one of the advantages of modern computing and communications supposed to be that we can build resilience through redundancy right into every layer of the stack, from paired PSUs and RAID through to hypervisors that automatically restart your processes on backup CPUs? Then use massive scaling of identical such units, and timesharing to ensure maximum utilisation, to bring down costs? With the multiple identical hardware boxes physically concentrated, allowing support and maintenance staff to be superbly expert in that specific kit rather than merely ok in handling a random assortment of parts? Thereby allowing all of the differences in the workloads to be concentrated into just the topmost layer, the customer applications, so that each can concentrate on what is unique and revenue generating, relying upon a robust infrastructure to supply the common parts?
Is everyone absolutely certain that the current arrangement (which is presumably this "Cloud" we keep hearing about) actually sensible?
If I may: https://nitter.net/elonmusk/status/1980221072512635117
Don't forget - Windows used to allow lots of colour schemes for Windows, even ones you could set up for yourself, then gradually de-emphasised and finally de-implemented them.
So, from Microsoft's point of view, your KDE is terribly, terribly behind the times.
Meanwhile, loyal Windows users are being treated to A While New Paradigm* ushered in by Ease Of Use features such as typing, veeeeerrrrry carefully "vivetool /enable /id:57857165,57994323,48433719,49453572,58383338,59270880,59203365" (you did get right, didn't you? Oh, you typed 718 instead of 719? Never mind, it is an easy fix; start with format c: )
* I'm assuming this is how they are talking about it, seems likely
And it looks like we've got an entire extra ten floating around at the start of TFA:
> prefer works written by AI to human human-written imitations
This is, if course, just Mr Claburn trying to make us *think* that this was written by AI because, as the piece points out, we prefer it that way. A cunning double-bluff and an illustration of how well fine-tuned the El Reg writing staff is.
The generic chippery still provides PS/2 so not making it available is the choice that is being made; plenty of current mainboards on sale today even bring them out to the rear panel (see comment above for a couple of URLs). As at least one other person has found (see other comments around here) even laptops can be using PS/2 - why bother with building up internal USB if the lines are there for the simpler alternative?
My two most recent PC mainboards, still very much on sale and advertised as Win11 compliant*, both have PS/2 sockets on the back: Asrock N100DC-ITX and N100M. No, I didn't go looking for that as a feature.
Asus "Prime" and "Corporate" mainboards all seem to provide PS/2. Harder to tell quickly with Supermicro, as their site is naff and doesn't hasn't images for the rear i/o, but from the top view, if that is the Ethernet, those USB then this other size of silver box, always in the same corner, is almost certainly PS/2.
In fact, it mostly seems to be "gaming" mainboards, and the grotty things that get crammed into name-brand boxes like HP, that are more likely to be missing them. On the outside, at least (there are still plenty of mainboard headers for serial ports, if one looks, so the missing PS/2 might be there as well - the generic chippery still provides the pins).
* not that they've been anywhere near Windows
By problems in the US. Hmm.
Time for reminders about "it is just someone else's computer", if it is in another country it's really not under your control (especially when "you" are a branch of government) and it is just hosting, not some specialised ability that isn't available elsewhere.
> toilet/smoke/coffee breaks? That's where the real time is lost.
Time is lost? Aren't those the times, like the proverbial "water cooler chats", when cross-fertilisation occurs* between team members, getting the chance to rub shoulders* with those outside your immediate cubicle mates, boasting camaraderie and increasing company value? Or just a few minutes of peace when the old noodle can do its magic.
From an engineering/development p.o.v. it's often seemed sensible to give your thinkers a number of paid timeslots for taking a good, long, shower, given how many times you hear "the answer just came to me as I was lathering up". Never convinced management of this, though.
* Okay, maybe a bad choice of words with regards to the first of those time-wasters.
These 46 minutes - how much is that worth to LBG, how much is it currently costing them in fees and, if(!) the fees change*, what is the cutoff when it is no longer worth pursuing?
You know, just some basic facts, the sort of bottom line entries that banks calculate all the time (?!) and they usually like to boast about, to demonstrate to the board/investors/mere customers that they've done the due diligence.
Or do they not have any real numbers? Maybe this is just something Copilot told them was happening?
* say, AI floggers feel like they'd like to makke a profit on the processing costs
> Most UPS manufacturers will charge you extra for the network card
Or the trivial piece of cable that will connect the USB data lines from the deliberately non-standard socket on the back of the cheaper units (hint: after you've found online the connections needed, when you spend the couple of minutes to make your own frankencable - it really is just a cable - connect ground and data but *don't* connect the +5V line on the USB, it probably isn't!).
They still have to solve the fundamental problem of distributed, massively multi-core processing used for a Papal Mainframe AI: which box do you put the pointy hat on?[1]
[1] the original reason for creating The Silence: one would stand by every cabinet, so that anyone who approached and realised that neither the hat nor the slippers were being worn, meaning that none of the AI's pronouncements could be deemed infallible, would turn from the machine room - and immediately forget what they had just seen.
> You'd like to think that there is at least one person who has got rich from tech willing to fund that.
Sure, all you have to do is find the person who got rich from tech that *didn't* (and doesn't) expect the user to keep buying new versions/models/upgrades/subscriptions and actually has real liquid money to use, not purely "on paper value, if we can find anyone daft enough to buy these stocks/options/massive debts".
A USB Hub with individual switches is a useful thing to have around, for a laptop or, especially, for a honking great tower PC: leave all the points of intrusion plugged in (no reaching round the back every time).
PS
Other retailers are available, Amazon is quite easy to search as a starting point
You remember the fun we had when Windows 3 programs could all see each others HWNDs? And you could hook any messages?
And we only used this for serious things, like debugging our own dialog boxes.
We *never* wrote anything that tracked the mouse and swapped the positions of the "Ok" and "Cancel" buttons whenever the pointer got too close to them.
It sees the man of the house[1] approaching a half-built bookcase with a screw in one hand: and the hammer runs away, hiding itself under the telly, the flat-head screwdriver follows closely[2], whilst the philips-head screwdriver keeps bumping itself up against his shoe.
[1] not our house, of course, *I* know what I'm doing, ahem[3]
[2] there being no paint tins in view
[3] so, where did I put the gaffer tape? That'll fix it - Oi! Come back here, you little blighter. Ha, got you cornered.
Don't suppose they could sort out the immobile coverage first? Like, say, anywhere near my house?
Still, if the trains all get coverage, maybe I could walk along to our station - it's only a couple of minutes away - and wait for the 09:45 if I want to send an SMS*
* don't want to go overboard, like hoping to talk to anyone
Let us first see if AI can actually do the first line roles entirely on their own for the next couple of yeras (let alone 5 to 10), without eating up *all* the time of the current second liners cleaning up after them - and whilst still remaining affordable, now that prices are starting to become more realistic.
> Bridenstein missed out the reason why NASA did not go with in orbit refuelling years ago: Senator Richard Shelby said he would slash their budget if they even mentioned the possibility of an orbital propellant depot.
Thanks for that; I'd missed the reporting on that, in 2019, but you prompted me to look it up.
For example: The SLS rocket may have curbed development of on-orbit refueling for a decade
> A moon base, according to various studies will have a mass of at minimum 150 tonnes (much more if you are trying out manufacturing or mining). For comparison, ISS is about 450 tonnes.
The answer is right there in the question: boost the ISS on a slow trajectory to the Moon, gentle it down and just be sure its solar panels are right side up before landing. The ISS is even already equipped with places to attach the boosters and the modular design will allow reconfiguration during the glide into place (there are some bits and bobs that you might want to move, as they are sticking out of the bottom).
There may be a few minor (cough) technical details to sort out. And apparently it is getting a bit whiffy.
Whilst the idea is intriguing, can't help feeling that it is going to be limited by the rate of churn in the mobile hardware market: by the time they have support for one device it'll have been declared "obsolete" by the manufacturer.
Whilst this may mean that someone who is really interested in running a Librephone system may be happy with the older model (possibly very happy, if they got a good second-hand deal) it will struggle to appear viable to the many people who insist on riding (close to) the leading edge of phone bling.
Ah, you are looking for the way the Bad Actors will attack the LLM, against the desires of Altman et al.
Whilst I'm looking at the way the LLM will be intended to attack (sorry, "help") you, for the benefit of Sam, the Worst Actor*.
* though he does manage to keep a straight face and rarely corpses in press meetings even when saying the most outrageous lines, so he has some acting skills
And as luck(?) would have it, a couple more El Reg stories later:
>> Japan tells OpenAI to stop spiriting away its copyrighted anime: "When I tried entering a prompt into Sora 2, it generated a succession of images of popular anime characters with such high quality that it was indistinguishable from the real thing," Shiozaki said in a blog post written days after Altman's Sora 2 note. "However, for some reason, characters whose rights are owned by major American companies, such as Mickey Mouse or Superman, did not appear."
So Supergirl is clearly out.
ChatGPT: "You've been mentioning Led Zeppelin often, would you say you like their music?"
Poor sap: "Oh, yes; I just wish I could listen to all their tracks."
... conversation continues, on and off, for another 90 minutes ...
ChatGPT: "You ought to get the door"
... doorbell rings ...
Poor sap: "Wow! How did you know there was someone at the door?"
ChatGPT: "Oh, that is the Amazon delivery; I ordered the full set of Led Zeppelin recordings for you, including the bootlegs. I wasn't sure what format you preferred, so I ordered the lot. Thank you for ordering from our Amazon affiliates page."
You are feeding ChatGPT's output back into itself?
You maniac!
Don't you realise that is just one step away from trying to google "Google" and destroying the Internet!
PS
Have you tried that experiment with an image whose description, when fed back in, won't ask ChatGPT to generate an image that clearly violates the intellectual property of a company that is big enough, and owns enough media outlets, that Altman ('s lawyers) actually gives a shit about?
PPS
Not that any of the above negates your final conclusion, btw, just, perhaps, one of the paths that seems to lead to it. Many other paths also exist.
> That so many people just accept its (M$) failures is.... beyond the pale.
That is being a *bit* harsh, isn't?
Saying it is beyond belief - oh, if only that were true. Beggars belief - closer to the money, particularly in the august company of El Reg. But, casting aspersions on the moral integrity of those who have fallen victim to M$ is perhaps too close to victim blaming for comfort.
Proper British units?
Wel, the spot price for a UK gold sovereign is just above £690 as I write this, and $35 comes out as £26.07 (hmm, that is about a florin low for a model 3B+ ex-VAT, which is your best choice for a pi-hole, but never mind). So you'll be getting 26 R'Pis to the sovereign, with a bit leftover for p&p.
Post being responded to: >>> 2025 people video themselves taking Tylenol and post it on the internet to show that it is safe cos someone the media told them to dislike the person who repeated the previously undisputed claim.
Corner's response: >> Nope. Person who took the above hints, still without any causation, and turned it into a strong claim, directly telling people that it was a cause of Autism
And now: > False. RFK Jr did not say this. Or even imply this. You have been fed a lie by the media and swallowed it whole.
Hmm. Here is a transcript of Trump's announcement from which one quotes "Acetaminophen. Acetaminophen. Is that okay? Which is basically, commonly known as Tylenol. Can be associated with a very increased risk of autism. So taking Tylenol is not good. I’ll say it, it’s not good." followed by the announcement of the FDA notice.
Sounds to me like a strong claim, by "someone the media told them to dislike".
Meanwhile, as you decided to bring up RFK jr out of nowhere, you'll enjoy his words from the same transcript: "Thank you, Mr. President. To meet the President’s challenge, I ordered HHS to launch unprecedented all agency effort to identify all cause of autism, including toxic and pharmaceutical exposures... Today, we are announcing two important findings from our autism work that are vital for parents to know as they make these decisions. First, HHS will act on acetaminophen".
Now, does that sound like he is talking about autism or not? And tying it into use of Tylenol? Do you see anywhere that he attempts to correct Trump or does he just wholeheartedly support Trump's words?
> You have been fed a lie by the media and swallowed it whole.
Maybe you'd like to reconsider - did you perchance mean to put the word "I" in place of the word "you" in that sentence?
I hate to sound like I'm defending these AI companies, but I'm afraid all you have done is remind us just how utterly, utterly shitty modern web pages are, at 1.5 to 2 MB.
The comparison to make is against plain old ASCII text, that is what is contained within things like the chatbot's "200K context window" and the content of those "messages": plain text requests for "write me a hello world in STOIC" then a plain text program listing as a response.
It will actually take you a fair old while to type in one of your "a 1 million char text file" examples, especially if you intend to make it more meaningful than just holding down the 'l' key.