We have gone from cruise control to cars being able to drive themselves quite well in about a decade. The last percentage points of reliability are of course the hardest, but that’s a tremendously pessimistic take.

Just pattern recognition it is not the neural net everyone assumes it is.

Tesla’s current iteration of self-driving is based on neural networks. Certainly the computer vision is; there’s no other way we have of doing computer vision that works at all well and, according to this article from last year it’s true for the decision-making too.

Of course, the whole task of self-driving is “pattern recognition”; neural networks are just one way of achieving that.

Indeed.

Comparing huge multinational countries which serve every country to the half of countries with the smallest energy usage is not terribly illustrative.

It’s a good idea to have some part of the benefits system make an attempt to work out whether people are cheating it. It’s a really bad idea to make that the focus of the whole system.

And my comment only used social media as an example. The point is, big websites have more draw than small websites, leading to a self-amplifying effect.

From the user’s perspective it’s not about “reach”; it’s about simply having people to interact with. If you go to a thread on reddit there’ll be hundreds or thousands of people to talk about it with, and there’ll be active communities for all kinds of niches. If you want to avoid reddit - whether because of privacy issues or site policy or mods or whatever - you have to deal with the fact that everyone else is sticking with reddit.

They analysis shows that, were the last election fought on the new constituency boundaries, the Conservative majority would be 14 seats larger. It also reveals that the swing to Labour required for the party to achieve an overall majority increases by 0.7 percentage points from 12.0 to 12.7.

So this is less representative. However, you’d be hard pushed to find anything that really looks like clear gerrymandering.

It’s like any number of blog hosts that have gone before it.

Absolutely true that this isn’t covered enough - the cuts were massive. They’ve also only been topped up by increases to regressive council tax, so it’s yet another way the Tories are worsening inequality through tax.

But the important question at the end - what the next, presumably Labour, government will do about it - seems unanswerable. It’s not just that Labour are now allergic to saying they’ll spend money. It’s also that paying debt interest is now a significant expense due to higher interest rates and our economy is barely growing, so taking on government debt no longer makes financial sense.

The Tories have done the opposite of fixing the roof while the sun was shining - they neither reduced public sector debt nor did they reinforce our services by investing cheap money into them. Now we’re left with the consequences and I don’t see how anyone can fix it.

Fair enough, I genuinely misread and thought that was within the quotation marks. But her message is still wrong because she is still talking about AI in general, but her argument applies only to a) AI whose data is derived from data scrapers like Facebook or b) AI put to surveillance tasks. That does not apply to Stable Diffusion, which is why I mentioned it, but it is caught by her assertion, “AI is a surveillance technology.”

AI absolutely has the potential to be used for surveillance; its use in facial recognition most obviously. But the person quoted in the article didn’t say “AI has the potential to be used for surveillance” - she said “AI is fundamentally a surveillance technology”. So if she’s not talking about LLMs and image generators, why is she saying that it’s a fundamental part of the technology? It’s not very fundamental if these two year-defining AI technologies aren’t included in it.

ChatGPT is not reliable enough to be worth citing. “Per ChatGPT” may as well be “per some bloke down the pub”.

Remarkably stupid take. It’s produced by big-data companies because you need a lot of data to feed it but that doesn’t make it “surveillance technology”. Stable Diffusion wasn’t trained on the kind of data she’s talking about, and it can’t be used to surveil you either. ChatGPT no more permits surveillance of its users than does chatting with a real person.

It’s spelled “copyright”, not “copywrite” btw :P

I know how HTTP works. These banners are supposed to (and are legally allowed to) store a cookie saying you have refused. Websites are allowed to store session cookies with displaying a banner at all.

No, they set a cookie to store it, but with a low retention period, so you get bugged again.

A start would be to require sites to remember non-consents for at least as long as they remember consents. Why do I have to be asked about cookies by every site every month?

Your browser can not save third party cookies, but it might break some sites. Some advertising situations allow the use of first-party cookies, and blocking first-party cookies will break most sites.

In either case you will still have to fill out the consent form, and if the consent is stored in the kind of storage you block, then you will have to fill it out every single time you visit.

Yes, but it often doesn’t work and even when it does the site is unusable while it works, which for some particularly awful banners is several minutes. The situation is worse on mobile where most people have a browser that you can’t install add-ons to (and I’m not sure if that one works in firefox mobile anyway)