* Posts by theOtherJT

953 publicly visible posts • joined 6 Jun 2013

Page:

Intel CEO suggests AI can help to create a one-person Unicorn

theOtherJT Silver badge

Here's how it will work. Go to Bangkok, go to Mexico city, go to Bangalore. Look at the city centre CBD where all the international big businesses are, and their lovely shiny buildings. Then drive out into the slums. See the people washing their one set of clothes in the open drain that passes for a water supply passing their tin shack.

That's us. That's all of us. Grinding abject poverty pressed up against incalculable wealth. That's the future economy. The very few ultra wealthy live in unbelievable luxury, and everyone else starves on their doorstep.

It happened in other places already, you bet your ass it can happen here too, and not only that but the ultra rich know this and are just fine with it.

Open sourcerers say suspected xz-style attacks continue to target maintainers

theOtherJT Silver badge

Re: The deeper issues

It's particularly hard to get this point across to developers who have never had to work in restricted environments.

A lot of us commentards are probably old enough to remember how precious every Kb of memory was and that you absolutely had to get your code to fit in X bytes at run time and that it could only be Y bytes on disk because there just wasn't enough space to do anything else. If your program wouldn't run in 640k of ram, then it wouldn't run, period. If it wouldn't fit on an 800K floppy, then you couldn't ship it because splitting it over two disks would massively increase production cost.

That's just not how a lot of current developers have ever had to work. They graduated from their software engineering degrees at a time when memory and disk space were already effectively infinite and have now got a decade of career experience behind them that's only re-enforced that fact - especially if they've been working "in the cloud" all that time, where you can just keep pulling more and more resource to make up for inefficient code.

In that world no one thinks it's weird that a single application might have 50 dependencies, which each have their own dependencies, or that the driver that changes the colour of lights on your mouse is 150M for some reason.

After delay due to xz, Ubuntu 24.04 'Noble Numbat' belatedly hits beta

theOtherJT Silver badge

Re: Minimal installation...

Surely you do what you always used to do and bring it up in 80 column text mode? The useful improvement would be to throw a "Hey, you took out the GPU I was using! Either put it back or fetch me a driver for the new one!" type message, which would be a very simple thing to do these days seeing as basically all drivers are sitting around in software repositories one way or another, and not a massive deal like it was in the 90s when we used to have to go and download megabytes of source over shoddy dial up lines and then spend an hour compiling the blasted things.

As to "make it impossible to read anything that's not English" Well... yes? I mean, I already can't read anything that's not in English because I'm an ignorant Brit who gets by on English, a few hundred words of French and a few dozen words of German - just enough for me to not get lost in either country when I drive through them to visit friends who live there. It feels like it shouldn't be beyond the bounds of possibility for these things to be apt/yum-able packages and if I don't need them I could just chose not to install them, but the very idea of an OS "Install" seems to have gone away in favour of "DD this image of a running system onto the device of your choosing - just so long as it's not a complicated device of any kind." which is kinda... irritating.

I get that this makes things simpler for people to use Linux on their Laptops etc, but lets be real - basically no one does that* - or at least those of us who do aren't bothered by having to read and select things from the install options when we do it.

Maybe I'm being unfair. Like I said - it's five or six gig and I've got five or six gig to spare and when I really don't have that I'll go find myself a Debian netinst image and start from there, but once again it feels like the pendulum of "simple" vs "useful" has swung too far toward simple in most distros.

*he says, from a laptop that's running Ubuntu 20.04 - but even that I installed by hand starting with the server distro and then bootstrapped my way up to a GUI so I could use a disk partitioning and encryption scheme that the desktop installer didn't seem to want me doing

theOtherJT Silver badge

Re: Minimal installation...

You know, I got somewhat curious so I decided to do a little experiment.

For entirely different reasons I found myself needing an Ubuntu machine that booted straight to a browser window - basically for digital signage, but we're going with Ubuntu as the base for a bunch of tedious policy reasons that I won't get into here. Firefox, as downloaded from their own website in deb form, and then installed onto a base version of Ubuntu server with just enough xorg on it that it's possible to target a single GUI application directly at the one running X session, comes out at just about 330M installed. In fact, from apt we see:

Need to get 80.2 MB of archives.

After this operation, 327 MB of additional disk space will be used.

A quick check of how big Firefox is running on my machine as a snap right now

du -sh /snap/firefox/current/

709M /snap/firefox/current/

Further inspection shows 222M of /snap/firefox/current/usr/share/hunspell which is a spellchecking library that's been included as part of the Firefox snap. I actually already have that installed on my machine in /usr/share/hunspell where it's taking up a grand total of 800K because I have only the single dictionary installed, seeing as I only really speak English. The Firefox snap has decided to include 138 others just in case I need them.

I'm sure that Firefox is far from the only example of this sort of thing so I don't want anyone to take this as me explicitly calling out Mozilla, but this pretty much encapsulates my problem with snap packages. I can't get rid of those dictionaries. They're utterly useless to me, but because they're pushed in as part of a larger package I'm stuck with them and even the one dictionary I do use is a useless duplication of one that's already in my systems /usr/share. 221M of wasted space right there.

Sure, the snap itself is a squashfs so it doesn't take up all that space on disk when it's not running, but in a way that's worse because now it has to waste cpu and memory duplicating things I don't need as well as disk space.

Every thing that's packaged in this way is going to be duplicating libraries and other shared-only-now-they-arent-shared type files to one extent or another. That sort of thing seriously adds up over time.

theOtherJT Silver badge

Re: Minimal installation...

This is true, but "Hey, at least it's not worse" isn't the most compelling argument against something.

I wouldn't mind it using a lot of space - if you can even call 5.6G a lot of space these days when even the cheapest of machines come with 256G SSDs in them - if I understood what the hell it was using all that space for. I'll trade space for features, or stability, or speed or... well, basically anything really, since I have a lot of space to spare but I cannot for the life of me work out what I'm actually buying with all the space I'm spending.

We never agreed to only buy HP ink, say printer owners

theOtherJT Silver badge

This feels like an own goal...

I can see this getting enough traction and generating negative news coverage to the point where people stop buying HP printers - even if HP win the actual case. You can imagine the thought process: Yeah, I think I heard some nebulous thing about HP breaking printers with software updates and there was a court case, not sure what that's all about, doesn't sound great, maybe I should buy epson or canon or dell or something instead. Seems not unlikely that they end up losing more money from customers buying elsewhere than they make out of lock-in.

AI will reduce workforce, say 41% of surveyed executives

theOtherJT Silver badge

Re: "Robotic engineers"???

I keep thinking of this wonderful exchange from one of the Fallout games where the local police have been replaced with robots:

Help!

robot: Stay calm, citizen. Watoga Emergency Services is here to assist you. Please state the nature of your emergency.

Mr. Sparkles is in the tree.

robot: Citizen, clarify. What is a "Mr. Sparkles?"

A cat. And he's in that tree!

robot: A cat is in the tree.

Yeah! He shouldn't be in there!

robot: The cat should not be in the tree. Is this correct, Citizen?

Yes! Hurry!

robot: Citizen, please stand back. <Sound of shooting and cat screaming>

No!!!

robot: Situation under control. Cat is no longer in tree.

Mr. Sparkles...

BOFH: So you want more boardroom tech that no one knows how to use

theOtherJT Silver badge

Re: I was once rushed to a meeting...

No, that was definitely an autocorrect, but I'll take it.

theOtherJT Silver badge

Re: Laminate everything

We tried the "laminate a sheet with the common options" but it was deemed "untidy" and the conference booking staff kept making us take it down.

See, the problem was that the main theatre had both a left and a right projector, the idea being that it could be used as a video conference room where you could put the screen showing all the other participants on one side, and the screen showing your presentation on the other side. It also came with a pair of PCs in it - a Linux one and a Windows one. We were a technical type institute, so a lot of people had fairly esoteric custom software and needed Linux to run it. We also had a socket in the lecturn to plug a laptop into, if you had some particular reason to bring your own.

Now we have the case where you can have:

Windows Screen1 left, Screen2 right.

Linux Screen1 left, Screen2 right.

Laptop left

Laptop right.

Windows left, Laptop right

Linux left, Laptop right

Linux left, Windows right

And then there was the "overflow" and "screen cast" features that let you use the other theatres as overflow seating, so you could _also_ decide to PiP any combination of the above and send it to the IP streaming device that would encode it and send it to one of the other rooms, or to either of the screens or - if you were feeling really fancy - have the AV techs up in the control room mix it with another stream from one of the half dozen cameras that captured different angles of the theatre:

Tight shot on lecturn for presenter,

wide shot of stage,

wide shot of audience,

regular stage focus left screen only

regular stage focus right screen only,

ultra wide from back of hall capturing audience and stage

These could also be sent to either projector and streamed to another room, and streamed to the internet.

The overall matrix of what inputs could be sent to what outputs numbered comfortably in the hundreds. The spectacular complexity of the thing necessitated a whole full-height rack in the projection room full of HDMI switches, mixers, amplifiers, video-to-ip encoders and decoders... It was absolutely impossible for anyone to use that room without an AV technician running it from the big mixing desk at the back.

Of course the booking team would very happily rent it to people without a technician on the grounds that "They don't need to use all that". Which of course they didn't, but given that it was there it just confused the absolute shit out of everyone.

We would have been much better served with a big button on the lecturn that just put whatever was showing on the machine there up on one of the projectors and turned everything else off. Since, you know, that's what everyone actually did with the room anyway.

theOtherJT Silver badge

I was once rushed to a meeting...

...where a contract worth tens of millions of pounds was being agreed. I was working for a large conference venue at the time which shall remain nameless but we had all sorts in over the years. This particular lot was compromised of various managing directors, members of government, academic think-tanks. Basically no one on less than 6 figures, and most much more than that.

The conference organiser was furious because the laptop we provided had "stopped working" and given the value of the conference this was completely unacceptable, out of order, we'll be speaking to your manager etc etc.

The problem turned out to be two fold. Firstly, our laptop (some kind of Dell Inspiron if I remember correctly) had been packed up and put in a cupboard by one of the conference guests because it was "untidy".

They had substituted it with their band new MacBook air, for which, amazingly, they had the correct dongle to connect it to the rooms projection system.

The second problem was that it was asleep. Has they done literally anything, pressed any key. Moved the mouse. Touched the track pad. Anything at all, it would have woken up. Instead, this room full of people whose collective annual compensation package had to be worth more than the building we were standing in panicked and called the conference organiser, who panicked and called IT.

Ad agency boss owned two Ferraris but wouldn't buy a real server

theOtherJT Silver badge

Re: Why two?

Because you want one mid engined V8 like a 488 or hooning around like a lunatic in and a big front engined V12 like a 812 for luxury grand touring.

Whizkids jimmy OpenAI, Google's closed models

theOtherJT Silver badge

In the words of Wanda Maximoff...

So. When openAI uses totally public endpoints to collect data that trains the model it's fair use, when security researchers use public endpoints to disclose openAIs inner workings it's a flaw that has to be patched immediately?

...that doesn't seem fair.

British Library pushes the cloud button, says legacy IT estate cause of hefty rebuild

theOtherJT Silver badge

Re: This will be unpopular but that is wrong with the cloud

Simple answer, from my perspective: it's outsourcing. Outsourcing always looks cheaper to start with, but ends up costing more long term. Ultimately you either own or rent, and renting is never, ever, cheaper in the long run.

theOtherJT Silver badge

Re: Shocked I tell ya !

...and I'm certain that going to the cloud will help with that.

On prem we ran our estate with 4 people. In the cloud we now need 11, and we have had to dump a ton of responsibility onto individual businesses units to manage their own infra and those new cloud people are getting paid more than the old sysadmins were.

Managing cloud is different. Sometimes better, sometimes worse, but almost always different and paying for people who know how to do it isn't cheap.

Microsoft calls AI privacy complaint 'doomsday hyperbole'

theOtherJT Silver badge

Re: the only way...

Laws only exist that can be enforced. They can claim the law applies to you, but if they have no way to enforce it, it doesn't. Corporations might eventually get slapped down hard enough to be forced to abide by some law or other but until that happens the only safe thing to assume is that they will continue to violate it. Even then let's be honest, they probably will. I've lost track of the number of stories on this very site of "Corporation gets fined approximately 4 minutes profit for being very naughty. Promises not to do it again." or "Government found to be in violation of it's own data protection rules. Agrees that this is unfortunate." I can't remember reading a story that said "Corporation forced to dissolve by bankrupting fine" or "Executives / government members jailed" in relation to the same.

theOtherJT Silver badge

Re: the only way...

Sure, but the internet doesn't exist only in Europe. Europe (or for that matter any other legislative body) can issue all the laws they want about the internet, but since it spans the entire planet by design you really can't trust that the specific place this particular bit of the internet is located gives a shit about laws where you happen to be.

It's something no government has managed to get it's head around. The internet makes international boundaries basically invisible and irrelevant. It's the unregulated wild west out there, and we can all pretend that laws apply to it but they just don't because anyone who doesn't like said law can simply host their service somewhere where that law doesn't exist. I mean do you check that every service you're using is hosted in a jurisdiction where your rights are respected?

theOtherJT Silver badge

the only way...

...not to surrender all of our personal information, family photographs, copyrighted works, art, and more would be to cease using the internet altogether.

Um... yes? Good. You finally get it. Were they not paying attention for the last 25 years? The Internet is a public space, and where it's not public, it's a space where you are permitted to exist by the companies that own it. It always has been this way. Unless you run your own server and set your own terms about who can access it there has never been any expectation of privacy here. We were telling people 25 years ago that you need to be careful what you say online and who you say it to because once something is "out there" it's basically impossible to get it back. This isn't new, it's just something that - for lack of a better term "regular" - people are waking up to now because it's in the news. We've known this since the beginning. The internet isn't safe. It's never been safe.

Linux for older phones postmarketOS changes its init system

theOtherJT Silver badge

Re: proof

I came in here to say exactly this, but you already did, so all I can give you is my upvote.

This, this right here, is why systemd pisses me off more than anything else. It gets on things and then you can't get it off no matter how much you soak it.

The S in IoT stands for security. You'll never secure all the Things

theOtherJT Silver badge

You are of course right that people don't care. That's not terribly surprising, because to most of them none of what we commentards would warn them about sounds like words in any language they speak. I've found it painfully obviously when talking to various members of my family about IoT things that they simply do not understand any of the things I'm telling them, and don't believe the things they do understand because "They wouldn't be allowed to sell this if that was true!"

I've had more than one family member tell me exactly that, and it really gets to me because that's exactly how this should work, but doesn't. There needs to be a law that says you - product vendor - are legally responsible for meeting these minimum security standards and if you don't you pay huge damages. People who are used to simple electronics that have a kite mark of some kind on them to prove that they're "safe" just assume that there's some equivalent for IoT devices, when in fact, there isn't. It's the only way that people who don't understand, and don't care to understand, the actual problem can be kept safe.

There's a reason that you find a little label saying "This device must not cause harmful interference" stuck on pretty much all electrical equipment these days. Someone had to legislate for that because until they did we had a bunch of cobbled together electronic crap out there that fucked up everyone's TV and radio signals as soon as it was turned on - but of course the people who bought it had no idea why and probably weren't even aware that it did. Legislation is the only answer here.

Year of Linux on the desktop creeps closer as market share rises a little

theOtherJT Silver badge

Re: Look At Cutting-Edge Enterprises ...

Most of the ones I know use Macs.

theOtherJT Silver badge

Re: Repeat after me:

Trust no one. No one can be trusted. It only takes one moment of carelessness or a single malicious employee to do catastrophic damage.

I have a user who has, because he has local root on a shared GPU cluster, managed to force us to send his entire team home multiple times because he's totally hosed the cluster by playing with things he doesn't understand to the point where we've had to actually go into the server room and rebuild the thing from the local console.

He's not a bad person, but he's absolutely not qualified to be administering a Linux system. He's a software developer. Presumably he's a good software developer, because despite the number of times he's done things like this and we've not let him go, but he can't be trusted with local root.

It only takes one fuckup to seriously ruin everyone's day.

theOtherJT Silver badge

Re: Repeat after me:

I totally take your point. We had full time 5 (it was meant to be 7, but for all the time I worked there, we never managed to have the complete set due to staff churn caused by the shockingly low pay given the complexity of the work) members of IT staff to support a company with about 1000 employees.

Personally, I think that's pretty impressive. That means that each one of us was handling all the support requirements of 200 people, and this was definitely only possible because 90% of those people were using Linux systems that were basically impossible for them to break - and if they did break, due to hardware failure or something, they were all effectively identical and we could just swap in one from spares and have them going again in a matter of minutes. Compare and contrast where I work now where we have effectively 14 proper IT staff and another 4 help-desk technicians to support about 400 people and still provide what I see as a much worse user experience, but can't do anything about it because our management absolutely insist that all users must get local root.

I think what @cyberdemon was getting at in their "Lazy IT" comment was that it's a matter of training the IT staff to be able to run a thing like this, and not just relying on people who are just about good enough at their jobs to click through the list of GPOs and it'll probably work, most of the time, for most people, maybe.

I'm not sure it's fair to call those people "Lazy" perhaps so much as "low skilled" and you do get what you pay for. It seems like the barrier to entry for managing Linux at scale is much higher than it is for Windows, but that itself has upsides as well as down. Basically any idiot (including myself, back when I was responsible for the Windows desktops) can click through the GUI and find a bunch of GPOs that sound about right and apply them to the fleet and get a "sort of good enough on a clear day" level of management, but it takes a much more skilled Windows specialist than I ever was to be on the ball about which of those were deprecated in the last release, which don't work for the "Education" edition but require "Enterprise", which just plain don't work any more... I got caught out by AD an awful lot of times. That's before we get into the problems caused by someone clicking though a UI rather than using something like Desired State Configuration (is that still a thing?) because a lot of the Windows admins I've met over the years don't even know that's something Windows can do.

But then again I'm not a Windows admin, and if that were my job full time I'd expect to be expected to know that sort of thing and design things accordingly. As a Unix admin I was required to know a very different set of things, and I don't actually think there's a massive difference in the number of good Unix and Windows admins out there, there's just a fuck ton of bad Windows admins available because of the ubiquity of Windows devices.

When it comes to managing Linux at scale the answer seems to be "You can do it properly or you can't do it at all." where as Windows is totally possible to do badly.

theOtherJT Silver badge

Re: It is the UI

I'd give you the full list, but I'm afraid I can't find it. It's a bit like the rules for the badges. I've definitely seen it, but I have trouble finding it any time someone asks about it. The things I know work are i b strike code a sub sup q ul li blockquote but there might be others.

theOtherJT Silver badge

Re: It is the UI

I was informed (in a way that I think was intended as an insult, but never mind) that for those of us with enough comment history to prove we're not bots that El Reg supports a small set of HTML tags. In this instance the one you want is code.

And yes, there's plenty of weird arcane shit in Windows, but Linux has a reputation for this stuff. Probably not really deserved at this point, but it's definitely there - especially amongst people who perhaps flirted with the thing a bit back in the early 2000s and were traumatized by being expected to occasionally do things in bash and swore "never again".

I personally love it, but bash is very much a product of a different era and to the average person it looks like fucking voodoo.

theOtherJT Silver badge

Re: Repeat after me:

| And what if you have 200 laptops scattered here, there and everywhere and need to make a configuration change?

I had a puppet agent running as root (which the users were not allowed to have) on the laptops with a public endpoint that it would poll for changes every 15 minutes. If I needed to change a machine I'd update the manifest and the next time the thing checked in it would update the machine config accordingly. This had the double benefit of the fact that if someone changed something I didn't want them to change but it wasn't practical to completely lock (Common ones like the VPN config where the username was something people might realistically need to update, but the host address absolutely was not) it would simply change it back.

It was, in fact, the same system we used to update the desktops and servers - even the switches by way of a switch management bastion. Everything was IaC. Everything was version controlled. All endpoints could be swapped from the "Prod" to "Test" branches as required. It even managed major version upgrades entirely remotely. We offered a "Things done fucked up" boot option that would connect to a sftp site, validate the machine certificate for your host, and download you the correct and pre-configured installed should something go terribly wrong. It was - in short - a metric fuck ton more reliable and convenient than AD was, where Microsoft had a nasty tendency to change what settings did between one version and the next, and occasionally just plain take them away from me.

Now, this was not cheap - at least in terms of man-hours spent getting the thing to that state. Basically the entire infra was totally custom and there were effectively two full-time members of staff to support it, but then I don't recall AD being particularly cheap in that respect either - although I imagine it would be easier to hire people for.

theOtherJT Silver badge

Re: It is the UI

Indeed. I'm typing this from a Dell XPS15 running Ubuntu 22.04. It's without a doubt the best machine I've ever had, but it was issued by the company I work for and it arrived in the office with Windows 10 on it, and we had to re-build it. Dell actually do offer Linux on the XPS13, but go to this page and look at the "operating systems" list on the left, it's not there. You have to select the machine with Windows and then customise it to chose Linux. What's more they only let you do that on one specific machine! Despite the fact that Ubuntu works perfectly on the XPS15, they won't sell it to you.

theOtherJT Silver badge

Re: Linux Mint

Sure, any distro can but steamOS is a product. It's a thing people can just buy and use without thinking about it. It changes the perception, and the perception of Linux has been more important than the reality for a long time. Linux has been "good enough" for a long time, but the perception doesn't match that. I personally like mint. It was my daily for a couple of years before I decided that stock Ubuntu was now good enough that I was happy to go back to that - but people really like a thing they can buy and don't have to think about. The major benefit is that it comes on the device and once it's gotten it's foot in the door that way...

theOtherJT Silver badge

Re: It is the UI

UIs are important, but TBH I think that Gnome has been a "good enough" desktop for quite a while now. Certainly for the last 4 years. Even my eighty-something year old parents are just fine with it. In fact they like it better than Windows because it has fewer buttons. They just pin the four or five apps they ever use to the Ubuntu taskbar and get on with their lives and it isn't constantly asking them to sign into things or do updates. Those I set to happen automatically in the background and they never thought about them again.

Here's why I think they wouldn't have bought machines with Ubuntu on if I'd not bought them for them:

They had no idea that such a thing existed.

And that is probably the biggest blocker to people taking up Linux as a desktop these days. Most people have either no clue there is such a thing, or if they do know about it, they think that's some nerd thing for nerdy nerds that they can't imagine themselves wanting because they're normal people who don't cast magic spells that look like this ${array[${#array[@]}-1]^^}.

Even for me - who is very aware that Linux exists and would very much like to be able to buy computers with it pre-installed - it's remarkably hard to do so. Most vendors will sell you Windows or Windows, and if you try and convince them to sell you a machine with no OS at all, because you don't want Windows, they just plain won't.

theOtherJT Silver badge

Re: Linux Mint

I'm going to have to disagree.

I reckon it'll be steamOS or whatever it is that Valves supports. Why do I think that? I think that gaming is a hook. Kids will own Steamdecks to play games, and then they'll get used to the idea that they can sort of use this thing as a PC too. Then as they get older they'll need an actual PC for work, but now - and here's the kicker - they're already comfortable with the idea of steamOS, and of course they still want to play games too, so why not stick with what they know?

It's the same way that the 8 bit micros completely dominated the 80s. Gaming drove adoption. Yeah, they got killed off in business by the IBM PC, but they didn't die in the home until the PC was also capable of playing games so it could supplant them and do both. Games is a really powerful driving force.

theOtherJT Silver badge

Re: Repeat after me:

Curious what it was that was missing in 2005, and what's still missing now?

The biggest blocker everywhere I've worked was always MS office. Specifically Outlook and Excel. My feelings about those two particular abominations and the way that so many office workers are suffering what I would best describe as Stockholm syndrome regards to them are well documented at this point, but those have been the big blockers for me.

I have worked somewhere where we very successfully managed nearly 1000 Linux desktops and that was the case as far back as 2012 - but even there, which I still remember extremely fondly because of how incredibly easy managing Linux desktops at that scale actually was - we had another 100 or so Windows desktops explicitly to provide MS office to specific members of staff who refused to couldn't work without it.

Anthropic unlocks Claude 3, claims it's better than ChatGPT and Gemini

theOtherJT Silver badge

Re: After spending the day with an LLM yesterday...

Well, quite. Hence the "scare quotes".

theOtherJT Silver badge

After spending the day with an LLM yesterday...

...I was genuinely impressed that it "understood" the question that I asked. That bit I am really blown away by. There's definitely room here for these things to be effectively a "better search" function for people who know what they want, but aren't sure of the exact terms they need to search for to get it. I particularly liked the fact that it linked to the sources it had used to generate the answers. When it comes to the content of the answers tho, things started to go a bit wrong.

Yes, they were obviously an attempt at answering the deliberately broad question I asked it, but they were not a correct answer. Multiple rounds of "No, that's wrong. You've made a mistake here." and it going "Oh, yes, you're right, that should read..." started out looking promising as it was fixing each mistake I pointed out, but each time it fixed something it introduced a new problem. After a dozen iterations it had re-introduced the error that I called it out on the first time, and when I pointed that out it didn't have any memory of the fact that it had screwed up in the first place.

So, yes. A useful tool in a lot of ways. It pulled out a ton of information related to my initial request, and gave me much better links to the things I needed to read to answer the question than a google search would have done, but for now at least - as in so many areas of life - it's still vital to check the original sources.

Cruise's valuation halved after its driverless car hit and dragged a woman

theOtherJT Silver badge

Almost nothing would be allowed...

...if we invented it today. That's one of the big problems with today. We've gotten so fucking comfortable with how well everything (looks like it) is going that we conveniently forget that a fuck ton of people had to struggle, and in a lot of cases die, for it. It's really easy to judge the mistakes of the past from up here, and IMO we're on a path already that leads to dangerous back-sliding.

It's a relatively short step from "We should not have done this thing in the past, it was dangerous." to "We should not do this thing it is dangerous." when we didn't live through how fucking awful the past was before we did the dangerous thing. Yeah, cars are pretty bad for us in a ton of ways, but you and I never had to shovel horse shit off the streets, or deal with the fact that the food we're going to eat today is a bit on the rotten side because it took a week to get it from where it was grown to where we bought it.

theOtherJT Silver badge

Re: "We've always sought to give users helpful, accurate, and unbiased information in our products"

You and I might be fortunate enough to see it that way, but ask any of the hundreds of people who get pulled over in the US* for "Driving while black" and I suspect they'll feel differently - and that's just the ones who got pulled over and then sent on their way.

Much like you skin colour doesn't really play a big part in my life so I don't tend to think about it much, but it is a big part of a lot of people's lives. We can debate endlessly that it shouldn't be, but if we're going to be honest we need to acknowledge that for a lot of people it is.

In that context history matters and correct historical representation matters. It's not OK to say "Everything is all fine and well with the world we live in now, no one thinks about things like that any more, isn't this a lovely civilized society we live in!" While ignoring the fact that for a lot of people, no it isn't, and for the vast majority of our history, no it wasn't. No problem ever got fixed by pretending it doesn't exist, and no problem that ever was fixed stayed that way if people started pretending it had never existed in the first place.

*and I imagine in a lot of other places, but literally every black person I know who has been to the US has gotten pulled over for as far as they could tell no reason at least once, and this is not true of all the white people I know.

Mamas, don't let your babies grow up to be coders, Jensen Huang warns

theOtherJT Silver badge

Re: Sure thing. All that's needed now...

Code. It's called code.

'How do I reset my router' isn't in LLM corpuses. An alliance of telcos wants to change that

theOtherJT Silver badge

I don't see this working...

...based on my - admittedly limited - understanding of LLMs and how they work.

So, as far as I understand it:

- To train an LLM it has to be able to make statistical inferences with a high degree of accuracy.

- To improve the accuracy of the statistical inference, you have to have a large dataset.

- The larger the dataset the more links the LLM can make between words and phrases and the higher the likelihood that some outcome is "correct" based on the fact that this outcome appears very often in the training model and other outcomes do not.

- Small datasets lead to weird behaviours from the LLM because there's not enough signal for them to start selecting for meaning instead of noise.

Given these facts (and feel free to correct me if we have any data-scientists in who know better) it's unsurprizing that if you ask big LLMs questions that are highly technical that have very few examples to draw from in the dataset they're likely to come up with silly answers. A fairly generic question like "how do I reset an *thing that needs resetting*" will probably have hundreds or thousands of examples of "You reset a *thing that needs resetting* by..." but if the thing is sufficiently rare it might have none at all for the specific thing you're asking for.

It doesn't seem to be in the nature of LLM's to go "Fuck if I know." but instead they tend to come out with something plausible sounding based on the data that is in their training set, but which actually bears no relation to reality.

So, ok, to try and fix that problem, we train the model on some very specific data that does contain all the information about your specialist field.

...but that's a small dataset. Which leads to weird behavior from the LLM because there's not enough in it for it to build useful statistical inferences.

Sounds to me like what they actually need is the operators manuals for the equipment they buy. Which I have to assume they have already, because they'll need it to train the model.

Perhaps they should try having their employees read them.

Google to reboot Gemini image gen in a few weeks after that anti-White race row

theOtherJT Silver badge

Developers must make a conscious effort to create models that produce diverse and fair results

Must they? Because this feels like another is/ought problem to me. What is the purpose of this model? Is it to reflect reality the way it actually is, or is it to depict reality as the people who designed it think it ought to be? When did we decide that the results of the model needed to be "diverse" or for that matter "fair" given that life itself is regularly neither?

I find the idea that we should "tune" a supposedly representative model to gloss over facts about life that we don't like really concerning. If there's a problem with the facts, focus on changing those facts. Don't pretend things aren't the way they are because they make you uncomfortable. I don't see any way that doing that doesn't make things worse rather than better.

Embedding the prejudices of the people who made the model into the model surely limit it's utility - and supposedly "positive" prejudice is still prejudice. The one thing these "AI" do is shine a lens on their training set. They show the things that are in there that we might not particularly like. It's about the only valuable thing I can see them doing at the moment, so deliberately stopping them from doing that strikes me as lopping off the one somewhat useful feature they have.

The machine is genuinely colour blind. It can't be sexist, or racist, or any other kind of ist because it's not aware of any such concepts. It's showing you the truth of what you showed it, warts and all. Once we start saying "You should tell us the truth all the time - but not that truth." we're taking the first well-intentioned step to a potentially pretty dark place because now I can't trust anything the AI says. What if you "tuned" it not to tell me other things you didn't find convenient?

Musk 'texts' Nadella about Windows 11's demands for a Microsoft account

theOtherJT Silver badge

Re: Pot, this is kettle

Because quite regularly some business who I'm invested in in one way or another - normally in as much as I use their product, or have to support people who use their product rather than a "I own shares in it" sense - announces something via bloody twitter, and then all the news outlets report it by linking to the bloody twitter, and I have no idea what the fuck they actually said because I don't have bloody twitter!

ChatGPT starts spouting nonsense in 'unexpected responses' shocker

theOtherJT Silver badge

He went on to state that...

...in reality, the systems have never been stable, and lack safety guarantees.

Never mind AI, that's the state of our entire fucking industry in one sentence right there.

Google debuts first Android 15 developer preview without a single mention of AI

theOtherJT Silver badge

Re: Word Definitions

In this case it's not you, it really is the children Google who are wrong.

Cutting kids off from the dark web – the solution can only ever be social

theOtherJT Silver badge

Children aren't idiots....

...well, ok, some children are idiots, but so are some adults. They're not idiots just because they're children.

Children are ignorant which is an entirely different thing, and ignorance is totally normal. Everyone is ignorant. There's too much to know not to be. Everything you know today, you were ignorant about until you learned about it. Telling kids they can't do something - as many people in this thread have already pointed out - is actually an amazing way to drive them to reduce their ignorance.

Sure, the dark web is a pretty sketchy place - but let's be fucking honest here, the regular web isn't exactly a well lit, soft-play area under constant supervision is it? The idea that we can achieve anything by trying to keep them ignorant is exactly the same kind of idiocy that the "If we don't tell them about sex, they won't have any" crowd perpetrates. It doesn't work. It's been proven over and over not to work.

For better or worse the internet exists. The dark web exists. You're not going to get anywhere by trying to keep kids from finding out about it or preventing them from using it. Like all potentially harmful things the best thing to do is explain to them what it is, what it's not, what can happen, and how best to protect yourself. Then you sit back and hope like fuck you have a good enough relationship with your child that they listen to you.

If you don't, there's no amount of law passing that's going to fix that one.

Google open sources file-identifying Magika AI for malware hunters and others

theOtherJT Silver badge

I've played Magika...

...I expect this to make about as much sense, but be significantly less fun.

It's time we add friction to digital experiences and slow them down

theOtherJT Silver badge

Re: Ain't Gonna Happen ...

Always my answer on the perpetual "What one thing could we ban to instantly make the world a better place?" question.

Sam Altman's chip ambitions may be loonier than feared

theOtherJT Silver badge

Re: I support him

Eh.... maybe.

There's always the wonderful 3rd option that we are both non-deterministic and non-free because the illusion of free will arises from the action of some effectively random process along the lines of atomic decay.

That being said I've always harboured a suspicion that atomic decay is not at all random - seeing as that it is at least statistically completely predictable, but we just lack the theoretical grounding to explain what causes it. If that's the case then it could just as well be the case with mental phenomena. Maybe they do actually "exist" in some physicalist sense, we just aren't looking in the right place.

I would immediately concede that this is pure speculation mind you, and anything I suggested to explain it without a theoretical basis that could be in some way tested might as well be defined as "Supernatural".

theOtherJT Silver badge

Re: I support him

"Do not" seems fair. I don't see any evidence of it.

"Cannot" however is a stretch. To quote(ish) Captain Jean Luc Picard:

"We took are machines, only of a different construction"

I don't see any reason we couldn't make a machine that thinks in theory.

BOFH: Hearken! The Shiny Button software speaks of Strategic Realignment

theOtherJT Silver badge

Shiny button software...

...As a service.

Missed a trick there. All shiny button software is a service these days. By paying a low monthly fee of $7 per user you can generate metrics that will allow you to waste at least 30 people's time at a rate of $50 an hour, for an hour per month while the compulsory meeting to discuss the metrics that no one has read takes place. The outcome of which will be a reduction in over-all spend by the equivalent of at most $1 per month per month, per user, by making everything slightly more annoying.

Billions lost to fraud and error during UK's pandemic spending spree

theOtherJT Silver badge

Officials working in counter-fraud told us...

"...the processes in place for sharing data, both between and within departmental groups, are often slow and burdensome, often resulting in incomplete or time-lagged data being shared"

And that, I can only assume, is entirely by design. The people responsible for setting these policies and processes would really rather that there wasn't a clear trail that would expose all the other incompetent decisions they made.

Joint European Torus experiments end on a 69 megajoules high

theOtherJT Silver badge
Pint

I've had the good fortune to see this thing in person...

...and it's absolutely amazing. Here's to you, you magnificent sciencey bastard. This one's for all those who worked on it. ->

Microsoft embraces its inner penguin as sudo sneaks into Windows 11

theOtherJT Silver badge

Good... I guess...

...but I have to wonder why now? What are they up to that this extremely obvious addition has finally made the cut after, what, three decades?

Says a lot about how I've come to view Microsoft that I view even what appears to be on the face of it good news with abject suspicion.

DEF CON is canceled! No, really this time – but the show will go on

theOtherJT Silver badge

Re: "That's because we studied math at school."

I guess it's because a lot of us don't understand where the fun is. Scratching some metallic foil off a piece of card doesn't on the face of it appear to be any fun.

I'm not anti-gambling in broad terms. I can get behind the idea of placing a bet on a football match - you were going to watch the match, now you get to watch the match and also maybe win some money. I play poker quite regularly.

I do not get lotteries which can be equated to "Pay a dollar and press this button. You have no control over the result. There might be money but there probably won't be." I don't understand slot machines for the same reason.

Page: