Posts by theOtherJT 1334 publicly visible posts • joined 6 Jun 2013
I had someone do that to me. I got out my phone, opened my IRC client, and pointed out all the messages that were being pushed in from my production CI system. Look at all my nice warnings. If any of these goes critical I'll get an alert right here. Same as your dashboard - except I didn't pay anyone to host it because this is all community code that's decades old operating according to standards that are even older.
I then SSH'd into my work gateway host and asked it to do a --dry-run update of all the infra. Nothing to do. See, all my infra is healthy, and if it wasn't I'd have been alerted and could have fixed it right here from my phone.
As long as we didn't lose internet in the office to the point at which I was unable to connect to it at all, I'm golden - and if we *had* lost internet in the office to that extent then all our cloud services would have become unavailable to anyone who wasn't working from home - which given that this was in 2016 was everyone, because basically no one was allowed back then.
Tell me what service I could have bought that would have been able to do that and how much I would have to pay for it please? Because this costs me precisely zero in licencing fees and is supported by the three in-house IT staff that we already had anyway who spend their spare time between doing desktop support, hardware replacements, and attending annoying meetings finding ways to make it more robust.
He decided he didn't want to talk to me any more.
God I miss working there. We're all cloud now at my current place.
You start to wonder if anyone has even heard the term "accountability costs". I mean, sure, ok, maybe we can save some money on some project by outsourcing it; but when it goes wrong - and it always goes wrong - what leverage do we have over the people responsible and can we get them to carry the costs for making it right? None? We can't? Oh. Not looking so much cheaper than doing it in-house now is it?
I'd have much more sympathy for this sort of thing if we negotiated contracts that said "You're getting paid £100 million a year for ten years to meet the following contractual terms. You will not get paid any more regardless of the cost over-runs. There will not be a contract extension, delivery on this timescale is non-negotiable. If you cannot meet all terms agreed at the pre-agreed milestones you will be considered in breach of contract and you will be required to re-pay everything that has been paid so far. If you go bankrupt because of this you will be nationalised, work will continue, and everyone at the executive level will be fired with no compensation."
Watch how fast people stop lowballing numbers and allowing massive over-runs when the terms of the contract actually include penalties that will hurt for not meeting them. Then watch how much the difference in cost between doing it your damn self and outsourcing it vanishes.
But this isn't like any of the others. Every time this has happened before it's been due to a sudden unexpected cut in supply - natural disasters, COVID, that sort of thing. The supply here has stayed pretty constant, it's just been totally redirected away from things people can actually buy into something very few people actually seem to want; namely massive rack servers to go in "AI" data centres.
So if the demand for memory from those data centres is the "new normal" so to speak, vendors should be investing in extra capacity. But they're not. So clearly they don't think that these "AI" bit-barns will still be generating demand in 5 years time. Why invest in the production capacity if you think the people who are currently driving up demand are all going to go bust and not be able to afford your product by the time all that extra capacity comes on line?
Feels like someone has made a very calculated decision on the near-term future of "AI" and decided they want to bet billions in additional business costs on it that may never see a return - which is refreshing, because it seems like everyone else can't throw money at this shit fast enough.
...because it would seem to me that their behaviour is going to be the thing to watch here. Are they increasing capacity? Clearly demand is up, so production capacity should go up to meet it if this is a long term trend - who doesn't want to sell more product to more customers and make more money after all.
If they're not, and as far as I can tell they aren't, it says something about how long they expect this surge in demand to last. It's going to take quite a few years to get a new chip fab up and off the ground, and cost a huge amount of money. If the memory vendors aren't investing in those I can only assume that they don't think that the demand will still be there by the time those new fabs would come online.
We all know that cutting edge IT kit has quite a short life cycle - especially under the kind of extreme stress you find in big data centres - so it would seem to me that if no one is investing in massive amounts more manufacturing capacity then they don't see a refresh cycle coming in 5 years or so... which is odd, because that's been a given for decades. Kit lasts 5 years. You replace it with newer kit that's more powerful / more efficient / both delete as appropriate.
Searle's argument is even weirder tbh. He's positing that a totally deterministic system can appear conscious despite having no consciousness in it - but then he goes on to claim that this proves that purely deterministic computational systems cannot possibly be conscious. Which... I mean I'm not expecting to find any consciousness in the atoms making up a brain either, but that doesn't therefore follow that the brain is not doing the thinking or that the mind that arises from it can't be described as conscious.
I've always been rather of the opinion that Searle was being deliberately contrarian with that paper and just dined out on how famous it got for the next 45 years so he wouldn't have to do any more work. ...which is of course the end objective of any good academic and one that I could only wish to emulate.
I once read a rather interesting paper on computational consciousness that describes such a system based on buckets of water on vast galaxy spanning belts capable of being emptied or filled in order to create an utterly gigantic universal Turing machine of the "each bucket is a cell, cells can be read or written containing precisely one byte each".
I believe it was by Daniel Dennett, although I read it over 20 years ago now and may be wrong. The point is that the complexity of the system can always be reduced to "input in, output out" and the "bigness" doesn't really enter into it. If that's the case we're not going to find consciousness by digging around in ever more complex systems because Turing already successfully proved that anything that can be computed at all can be computed on a UTM. Since LLM's are clearly performing computation if there *is* such a thing as consciousness going on in there, we're not going to find it in the structure itself - which could be arbitrarily redesigned to include some utterly bizarre machines without altering the result of the computation.
Look, consciousness is a thorny problem. But then so is everything if you want to get into the weeds. Philosophically you can just about "prove" the axiom "There are thoughts." Even "I think therefore I am" is problematic because it presumes a distinct concept of self separate from the thought.
There are arguments to be made that consciousness isn't even a real thing, and that qualia are some sort of emergent phenomenon that exist only so much as a ship does - namely because we say they do.
...and yes, I went there. Ships don't exist, belonging to Theseus or otherwise. Ships are just labels we stick on collections of atoms, which themselves are labels we ascribe to collections of protons neutrons and electrons, which themselves are only collections of... and down and down we go. Maybe there's a most fundamental particle down there somewhere, but hell if we know what it is. Everything is just convenient labels because we don't have the capacity to deal with un-abstracted reality.
It all gets incredibly tedious terribly quickly, take it from one who spent 4 years getting a degree in this shit.
So how do you prove consciousness? You don't. There isn't a test for it, and there necessarily can't be because we can't even properly define it in ways that aren't circularly referential.
You pretty much have to treat the word "Conscious" like you treat the word "Pretty". You're not going to go out into the world and grind it up into constituent parts and sieve out the particles of attractiveness whereby something can have more or less of them, it's just a word that exists because we mostly agree on what it means not because it has a formal definition. I chose "Pretty" quite deliberately because across cultures and even individuals there can be really quite different opinions on that.
That's because they've spent the last decade monetizing the shit out of the search engines to the point that they've become close to useless. Google especially is terrible compared to how it used to be. To abuse a quote: they're not making religion cool, they're just making rock and roll worse.
I mean, are we going to get a news story next week about not being able to get beer, or potatoes, or cutlery or some other totally banal thing because of the utter bullshit that AI is causing across all the supply chains?
I despise AI at this point.
"Why anybody lets these things anywhere near anything important, let alone "trusts" them with it is frankly mind-boggling."
Because the vast majority of people have absolutely no idea how they work, but have certainly heard lots of great advertisements for them from the snake-oil salesmen telling them how great they are and how much good they can do.
One day, I would really like to see some consumer protection legislation related lawsuits here. Products sold must be "Fit for purpose" and I think a lot of the sales push on these things suggests that they are capable of a wide variety of purposes for which they are absolute not fit.
...do we not think that we're attempting to close a door long after this particular horse has left the building?
I mean, sure, app stores should have content moderation - we're all well aware that they don't they just have algorithms that sort of work as long as no one looks too hard or pays a lot of money - but that aside there are dozens of websites doing this right now that aren't gated behind any kind of "store". The time to do something about this was waaaaaay back that way but everyone in charge was too excited about all the money they could save because that cheap "AI" could replace expensive people to take the time to question if there were going to be any negative consequences here.
To add a serious and depressing answer: because they want to own slaves. They want workers who have no choice but to work because they are owned by the company and they won't have to pay them.
This, ultimately, is what AGI is about. An intelligence you can own. No pesky salary to pay. No pension. No time off. Free intellectual labour, forever. That's why they want this. No more, no less.
"WE WANT TO BUY YOUR RAM! Has your business just completed a cloud transition? Do you have DDR4 or DDR5 ECC RAM? Due to market pressure we have HIGH DEMAND for memory and will BUY YOUR RAM!"
This is a real thing that actually happened. As far as I can tell they're legit too. They are a real company that actually exists and actually buys depreciated IT kit to sell on. According to their price list I have several thousand quid of memory in my desk drawers at the moment... I wonder if I can convince the bean-counters to let me offload it.
Unfortunately most of them are so willfully ignorant of everything going on in the world that they only come to that realisation when they personally lose their job. Hear about it happening to someone in another zip code and it's "fake news" so any change that's coming won't be coming quickly.
...we all lose. If it's the AI techbro crowd we can look forward to yet more slop everywhere. If it's the copyright lawyers it sets yet more precedent for them to sue your kids school because a 10 year old drew a picture of the latest Disney princess in an art class.
The only winners here will be the lawyers, as usual.
IMO the "apology" is actually worse than the error. The error shows that the thing isn't as capable as the user might hope it is. The apology is a deliberate lie. It's not sorry, it's not capable of being sorry. Whoever designed it to apologise like this is actively perpetuating that lie.
You know what, I've read it again, and I think you're probably right. They're not claiming that the stock management system was working only 95% of the time, but stock was available 95% of the time.
OK, that's more forgivable. There are definitely other ways that you can end up out of stock - sometimes the start align unexpectedly and everyone decides they want one specific thing on one specific day and there's nothing you can do about it because there was no way to know in advance that you'd need 90% increase over normal stock levels to cover demand that day.
That being said - they're a supermarket. Having things in stock is kinda their purpose, and a 95% hit rate still isn't that great.
Maybe I'm wearing rose tinted specs here, but anecdotally I don't remember going shopping 20 years ago and having to go to three different supermarkets to get my weekly in - something that happens to me quite regularly these days.
"Asda said on Friday that the completion of the system cutover disrupted operations during Q3, particularly the flow of stock between depots and stores, causing inconsistent availability levels across stores and particularly online."
Which they attribute to the failure of their stock management system.
...95 percent.
Ninety five.
How has it come to this? Seriously, how have numbers like this become acceptable? That's a 1 in 20 failure rate. That's shocking... or would be had that not been something I've come to expect these days. Can you imagine putting up with that in any other walk of life? Every three weeks your fridge decides to take a day off and ruins all your milk. One working day a month you just can't get there because your car has decided that this is that one day in twenty where the starter motor is having unscheduled down-time. Every twentieth trip to the shops - oops, sorry, can't sell you anything right now, it's the 5% of the year when the payment processing isn't working.
Why the hell do we put up with IT systems that are so bad?
Not going to get one of those tho, are we? Because let's be honest even the idiots at Microsoft know in their heart of hearts that that is the version that everyone actually wants, and the sales numbers would sink their "Everyone loves us for our excellent feature set and the massive value all our AI products add!" line by the end of the first quarter as everyone stopped buying the full-fat version.
...that people like Dave here aren't valued any more. I'm not quite as old as Dave, but even in my 20ish year career* it's become very obvious that the older guys have been very deliberately replaced by people who won't argue with management when they decide to do something dumb for the sake of "shareholder value" or to chase whatever the latest fad is.
The ones who were there back in the day, when the company was first starting out and had sufficient seniority to turn around to their project manager and say "No, we're not doing that, it's stupid" and get away with it have been ground down over time until the only ones left are the burnt out just-coming-in-for-the-paycheque types who don't care any more, and the younger grads who either don't know any better, or are too afraid of upsetting their promotion prospects to speak out.
* Not at Microsoft, but the same thing has happened in every private business I've worked for.
Unless you're looking for rackable gear I wouldn't bank on it. The half dozen ML servers I have sitting in my build room waiting for the finance team to finally agree that they've depreciated to the point where we can throw them out* are most definitely not the sort of thing you'd want in your home office, and the new ones are even more specialised.
The GPUs rely on case airflow for cooling, the mainboards have so much random shit in their bioses that they take an age of middle earth to boot, and all the CPUs are dog slow but massively multicore because that's useful for orchestration to the GPUs, which is what they were for - oh, and obviously they're all full of ECC ram which most consumer boards won't take. It'd be pretty hard to repurpose them to anything end-user facing.
* We did a cloud migration thing, and now we're not allowed to use any on-prem iron any more because otherwise we're not "Getting best value" from the cloud, but the on-prem machines still have a book value too high to get rid of, but not high enough for one of the finance team to take the time out of their busy schedule to establish if they have any actual resale value and try and move them on, so there they sit...
...of things I wish could be un-invented, knocking off the previous title holder of those automated phone lines where you can press 1 through 4 to do a variety of things that are nothing to do with what you want, or hold to be transferred to a different call centre who will disconnect you while transferring you to another department.
This is true, but for the sake of taking 5 minutes to fill out a bug report, I'll take that as an acceptable price to pay for getting a free operating system that doesn't advertise at me or report on everything I do.
...also linux bug hunting is kinda my day-job, so a lot of the time it's a twofer.
You should be tho. It doesn't matter that it doesn't work, the execs who decide who to make redundant have drunk the kool-aid and are going to be making those cuts anyway.
Sure, they might be re-hiring in a few years when it turns out that "Oh shit, we've made a terrible mistake" but I wouldn't bank on it. More likely they'll just bury their heads in the sand, and try and gas light everyone into believing that everything has always been this bad, and we all have to live with terrible service that can't answer simple questions for a few more years until the whole thing finally collapses in on itself due to being totally unaffordable and we get a colossal stock market crash leading to even more layoffs.
This will end badly.
Note that systemd compiled with musl has various limitations: since NSS or equivalent functionality is not available, nss-systemd, nss-resolve, DynamicUser=, systemd-homed, systemd-userdbd, the foreign UID ID, unprivileged systemd-nspawn, systemd-nsresourced, and so on will not work.
Half that shit doesn't work properly now.
Most people don't know that you can boot of a USB disk. I'd take a bet that most windows users don't actually know that it's possible to change operating systems. You buy a Mac, you get a Mac. You buy a PC you get Windows. What do you mean "Linux"?
There are far fewer of us than we think that know anything about how all this works.
