Posts by Paul 195 349 publicly visible posts • joined 20 Jan 2011
Your choice of language is quite revealing of your prejudices. We get it. You don't believe that CO2 is a problem or that renewable energy is viable. Because god forbid that we should use energy sources that don't pollute or make us dependent on importing most of it from elsewhere. What a relief it must have been to you that war in Ukraine didn't cause everyone's energy bills to shoot up overnight.... oh wait...
Building nuclear power stations is a terrible option. They are expensive, take a long time to build, we don't know how to dispose of the waste material, and in terms of cost/kw are a long way behind solar which gets cheaper every year. The way to make renewables work as a viable and resilient energy option are:
1) Over supply on the generation side - the sun doesn't always shine, but the wind is always blowing somewhere even if it isn't blowing in a particular locality.
2) A much better grid so that power can be shifted from where it's being generated to where it's needed.
3) More storage capacity to smooth out peaks in generation vs peaks in demand. A smart grid and electric cars could actually help here as each car plugged in is a mobile power storage unit.
And the other thing to do, is to stop wasting energy on useless shit like crypto currencies and building ever larger LLMs with dubious benefits.
So you are complaining about the BBC's inaccuracy by linking to the page where they list the errors they are aware of and correct them? Wouldn't it actually be quite nice if all news publishers provided this kind of aggregated detail about incorrect information, rather than reluctantly printing apologies in small print on page 9 in the hope no-one will notice?
"The CCP model"? AI is the capitalist's ultimate wet dream. No need to pay labour with its irritating demands for fair treatment and a share of the profits it helped create. Just ranks of obedient machines working without breaks, health care, or complaint. Exactly who is going to pay for the output once you've sacked all the peons is still kind of unclear. Fortunately for us, AI is so far of limited use for solving real world problems. The things it can do usefully and well aren't profitable enough to justify the vast sums of money being thrown at it; the things that would be profitable (replacing humans) are so far to risky given the general flakiness of generative AI. At some point the bubble will burst and it will make the Web 1.0 collapse ln 2001 ook small in comparison.
Plenty of women are interested in careers in technology but have such an awful time from men who can't believe a woman can be technically competent that they switch career direction. My spouse is one such. Admitttedly, her career as a real-time software dev started in 1989, but a lot of comment threads on el Reg lead me to believe that things haven't improved massively in the interim.
Having worked for a few years at a company that took the idea of diversity in hiring and promotion seriously, rather than just paying it lip service, I can tell you that I worked with several female engineers who were very talented, and because women have to work so hard to get taken seriously in software engineering roles I will say hand on heart that as a percentage I've met more mediocre and or incompetent male engineers than I have female ones.
Or at least it would be if they could make it work, then they could get rid of all those irritating workers who want to be paid for the time, have holidays, be looked after when they're sick, and so on.
Fortunately, they can't make it work, which is just as well because the dystopian future where a handful of people have everything and the rest of the world has nothing will end in tears for everyone, oligarchs included.
There are actually well-established rules under international law, including those laid down by the Geneva Convention. If there were no rules, and no laws pertaining to the conduct of combatants in war, no-one would ever stand trial or be found guilty of war crimes. Israel already stands accused of a number of war crimes in relation to its recent conduct in Gaza, and the IHRC are also considering the charge of genocide.
The reason for these laws are to establish bounds to ensure the safe treatment of prisoners of war, and to limit atrocities against civilians. It's also worth pointing out that Israel isn't actually at war with Lebanon. The British government was rightly outraged when Putin had two of his political opponents assassinated in this country. Imagine the upset if he'd detonated hundreds of remote control radio devices to kill people he didn't like who were based in the UK. Israel's conduct is not only barbaric, but is also not for any serious long-term strategic purpose. It's to prolong and widen a war that is the only thing between Netanyahu being put in jail by his own people.
Blowing up pagers is indiscriminate because you have no idea who will be nearby when you blow it up. You don't even know that your target will be holding it at the time. Which is why several children are reported casualties.
We can only hope this is added to the pile of war crimes that Netanyahu and his vile cronies are already accused of.
It looks like self-driving cars are better at avoiding the kind of fender benders that happen in traffic where one vehicle runs into the back of another. That's great, but those are also the accidents that generally carry low risk of death or injury. It looks like they are far worse at avoiding the kind of accidents in more complex scenarios, and those are often the ones where someone gets hurt. Nearly twice as bad as people according to the research.
Please don't let these things loose on the roads in British and European towns and cities which are full of "complex scenarios" and far harder to navigate than typical US style grids where pedestrians are only legally allowed to cross the road in certain places. The accumulation of training data has already cost several lives in the US.
Over the last 20 years "agile" has come to mean any number of nonsense things, none of which lead to agility, and many of which lead to tears before bedtime. Time and again organizations adopt some bastardised form of scrum with certified agile practitioners to help them. Far from not creating requirements before starting the project, requirements are defined in advance, the end point is fixed, and then a dev team go through this charade of fixed length sprints of between 2 and 4 weeks long.
Stories are not broken down into small enough pieces, the room to change direction is limited, and the delivery increments are too big. The methodology pushed on the dev teams is too restrictive, retros aren't taken seriously, and (one of the things identified by this study), there isn't real psychological safety so that teams can challenge practices or directions that are failing.
See this article for more: https://tanzu.vmware.com/content/blog/why-your-agile-initiative-is-failing
We already have access to a working fusion power generator* that costs us nothing. And the technology to turn its output into usable electric current has been getting steadily cheaper every year for a while now. Although the idea of cheap limitless energy is appealing, it's still looking like quite a long shot; unlike most engineering problems it seems that scaling down is the problem not scaling up. The biggest problem with renewables is energy storage for when the sun isn't shining or the wind isn't blowing; what kind of excellent energy storage technologies could be developed for a fraction of the investment in chasing the fusion unicorn?
* yes, I mean the sun.
"The people have spoken"
That's generally how populist tyrants justify themselves. Here's a little reminder of the electoral statistics for Brexit:
48% against
52% in favour
on a 72% turnout. Which means 38% of the people of voting age at the time of the referendum voted in favour. Hardly a ringing endorsement from "the people" there. Subsequent polling suggests that 1 in 5 of those who voted leave now regret their decision.
We have yet to see a single concrete benefit from Brexit. Jacob Rees-Mogg when made minister of Brexit Benefits was unable to articulate anything beyond high wattage vacuum cleaners. Meanwhile, our rivers and coastal waters are now literally full of shit because the government changed water regulations due to the difficulty of obtaining purification chemicals in the post-Brexit regime.
An aviation expert will correct me if I'm wrong, but didn't some of the early airbus planes lean too heavily into automation and they had to dial it back for causing the same problems as we see with Tesla software - it did too much causing pilots to disengage so that they then had problems resuming control when needed.
Software like Tesla autopilot is guaranteed to be dangerous. It works most of the time so drivers disengage. And then when it says "hey, I don't know what to do, help me out human", the human hasn't been maintaining context and hasn't got time to rebuild it.
The problem with this suggestion is that it assumes that all the self-driving vehicles only have to cope with other cars. But in the real world they need to cope with pedestrians, stray dogs, runaway prams etc. And the other problem with this suggestion is that lobbyists will argue that people, not cars, are the problem and we should have strict laws and segregation keeping them away from motor traffic.Which would make our cities even worse hellscapes than they already are.
If that seems improbable, here is your reminder that the US has jay-walking laws, legislating against people being able to cross roads, because of lobbying by motor manufacturers.
You've absolutely put your finger on the problem with self-driving software; it can't imagine or predict. A good many years ago on a country road I saw what looked like a couple of firelfies up ahead (it was a pitch-black night, no street-lighting or moonlight). The weird cyclic motion confused me though until I came up with the accurate hypothesis that I could see the reflectors on pedals for a bicycle being ridden with no lights. I slowed down and was able to pass said cyclist safely. Presumably an FSD Tesla would have ploughed into him at 50 mph.
And then legions of fan boys would have said the cyclist deserved it for not having any lights. Well, doubtless the cyclist has some responsibility for his own safety, but equally car owners have a responsibility not to trust control of their death machine to a half-witted piece of software demonstrably not up to the task.
Touchscreen controls are the worst. I've had several rentals where the heating/cooling controls were only accessible by touch screen, Which is not where you want them when you are hurtling down the autobahn and you've realised that you need more heating. They're a nasty cost-saving measure and objectively less safe than old fashioned knobs, switches and levers which you can work by touch alone.
None of which stopped another Tesla fanboy on a social media thread I was in from insisting that anyone who objected to them was a luddite who didn't understand technology.
I've got Tesla's new marketing slogan:
I came for the car, but I stayed for the cult.
A few years back we were always being told that self-driving cars were just around the corner (ho ho). I was sceptical then because it looked to me like all the successful testing was being done in relatively simple environments, and nothing as complicated as say London or Paris. For one thing, in busy areas with a mix of cars and pedestrians, the humans in the process will negotiate the space by making eye contact and using other non-verbal cues. Machines can't do that yet.
For self-driving vehicles to be viable, you would have to further restrict freedoms for pedestrians, and manufacturers would lobby for exactly that.
In fact history would repeat itself: https://www.bbc.co.uk/news/magazine-26073797 . I drive, but I find the idea that people's freedom in urban environments should be curtailed for the convenience of the car lobby rather dystopian.
I learned to drive manual, did it that way for 20 years but eventually switched to automatic after owning a manual Saab with a heavy clutch that was tiring to drive in heavy traffic. Manual gearboxes are not an option if you are driving hybrid (which I am these days) or electric. I don't miss manual gearboxes and when I hired a van with manual gears recently spent an upsetting three minutes trying to make the f***ing thing go into reverse as every manufacturer does that just differently enough to make it confusing. In this case the trick was liffting up a neat little ring partway down the gearstick.
For most of my life, experts have been saying that we are "about 20 years away from AI". In that time, we've seen a number of goals achieved: "play chess as well a a person", "play go as well as a person", "recognize a picture of a cat" and in every single case, it turns out that AI (or AGI is it is now called) remains as far away as ever.
Lots of excitement last year when it looked like Chat GPT and the cohort of LLMs could finally pass a Turing test. And yet, it looks like AI is as far away as ever. The biggest advance would appear to be that we don't understand exactly how the new models work any better than we understand how our own cognition works. So, yay, the experts have built systems they don't understand and can't predict.
The biggest clue that LLMs don't replicate human intelligence is the wild disparity in power consumption for both training and running the models. The human brain works its miracles on less than 100W. Good luck doing anything in an LLM with that kind of power draw.
With the confidence of someone who is not an expert, I predict that LLMs will be another blind alley in the search to replicate human level intelligence, although they look like they will have a number of useful applications.
I've been a developer working for a software vendor producing IDEs (it wasn't Jetbrains BTW). And I've also spent the last few years as a consultant developer working with enterprises to produce the sort of business software that the world runs on.
Those two groups developers view the world through very different lenses. The developers in the first group are smart talented people who like to do interesting things and are always trying to find ways to make their product better. But they live in an ivory tower, they aren't allowed to meet customers often, and sometimes they can run off in weird directions. The second group of developers have a mess of bureaucracy, process, and governance to deal with. Their world is very different.
To the first group of devs, sticking a bit of AI in there gives them fun work to do, and they think of all the ways it might make their own job easier.
For the second group of devs, as the article points out, the AI is a potential governance nightmare.
All I'm saying is, don't just blame management for this one.
Moore's law is no longer reliably delivering exponential performance increases from hardware, but the amount of stuff we want to run on computers goes up every year. At this point moving back from the abstract machine runtimes of C# and Java to native code looks like a savvy move to slow down datacenter's ceaseless appetite for electricity. I expect Rust and Golang to start creeping up the language charts as enterprises realise they can do more with less by switching to "faster" languages. GraalVM is a convincing attempt to do the same for the Java world.
I think the wider point is not whether this example is useful or not, it's that the protections built into LLMs are so easily bypassed. Putting these systems to work in the real world brings a whole category of brand new, hard to mitigate, set of vulnerabilities to software. None of these have made the OWASP top ten yet, but they are not going to be as easily fixed as perennial favourites like SQL injection and CSRF.
"1000s of terminals" are no longer a thing. These days nearly every large system actually runs over the web (not necessarily the public one), meaning that there are no terminals or dedicated network to maintain. Your delivery channel is just part of the regular IT infrastructure for any business. This usually applies even when the backend is IBM and CICS. You can deliver a green screen on a PC, but usually you've replaced it wth a prettified front end delivered over https.
Elon thinks he can ignore laws because he is special.
When someone who isn't a billionaire does this we call them a psychopath and lock them up.
Apparently, when Elon does this hundreds of fan boys rush to say he's great and everyone should do what he says.
Newsflash for fanboys: Elon doesn't care about you anymore than he cares about his workers' well-being.
Much of the Covid money was spent with newly minted companies that had no experience in the area of PPE, but did have access to the "VIP lane" of government ministers. While companies who were *actually in the business of making and supplying PPE* were ignored. I refer you to the case of Michelle Mone and PPE Medpro, the best known and most egregious case of this fraud. https://www.theguardian.com/uk-news/2023/nov/06/michelle-mone-admits-involvement-with-vip-lane-ppe-company
They are not all uber-rational financial and technical geniuses, anymore than everyone who works in government is a half-wit. I've seen the insides of enough well-respected titans of the FT100 to know that the "private sector" can be just as dysfunctional and slow on the uptake as any Whitehall department. BTW, neither the internet nor the world-wide web was created by the "private sector".
You don't know always know whether an innovation is actually feasible or a good idea till you try to build it. Sometimes the private sector puts up that cash, sometimes the government does. I can see from the screeds of information provided in the thread that this is one of those ideas that might/might not work, depending on the practicalities. So maybe it's worth taking a small scale punt on it to see if can be made to work as if it could there would be worthwhile benefits. And if it can't, we can put it back into the drawer with all the other stuff that never turned out to be practical.
There's always an assumption that the machine reacted quicker than a human driver would have done, but has that actually been validated? After all, self-driving software is a lot more complex than the motion detector that opens a supermarket door. And might a human driver have swerved and been able to mitigate the effect of hitting the pedestrian?
It's very hard to find accurate comparisons on accidents per mile for humans vs autonomous vehicles, and we tend to assume that because so many human drivers can be distracted, inattentive, or just plain terrible, that self-driving vehicles are better. But where's the actual evidence for it?
>> Making a lot of noise about google picking up that one is searching information about drilling appliances and slapping a few ads, is petty.
That's not all google is doing though. They track as many aspects of user behaviour as they can across as many devices as they can, and then combine that data to build up a profile. It's invasive behaviour, and they are very secretive about what they are doing. GDPR etc don't go far enough. The internet giants need to be subject to the same level of scrutiny that they apply to the population, and fined every time they use data for a purpose that users haven't explicitly agreed to.
You can sell advertising on the internet, and target it pretty well too, without needing the huge data warehouses Google, Meta, Microsoft, Apple, etc have built.
There's a lot to be said for keeping your baseline infrastructure in-house to contain costs, but using the hyperscalers for... when you need to scale up in a hurry. It's also easier to do new development and run experiments on a hyperscaler because you can provision what you need and then switch it off when you're finished. But it does, as you say, require discipline. And education. Left to their own devices users will provision large and expensive compute instances they don't need, and then leave them running for months.
>> Sounds like make-work to me. Has anybody actually done a cost analysis? I have.
No you haven't. To have any validity, such an analysis would require collecting a lot of data from different enterprises, and you would need to be able to calculate the value had been generated by work done with and without collaboration tools. It's the kind of analysis that would require.... wait for it.... a team to do. And you'd need to be able to charm all those different enterprises into giving you valuable and commercially sensitive information. Stop saying stuff to make yourself seem clever. It's having the opposite effect at this point.
@jake -
I work a lot with distributed teams, we run collaborative workshops, we do pair-programming, we work on architecture design, all sorts of things that require screen sharing, and where it's helpful to comunication to see the person you are talking to. You can't do any of those things on a voice only telephone. If your job can be done using voice only, possibly only talking to one other party, good for you.
Just stop parading your lack of imagination about other people's jobs.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
