That seems very unlikely, given that this blunder strengthens the factions that's opposed to regulation.
1243 posts • joined 24 Nov 2007
I used to try to make my docs as foolproof as possible, until I realized that nobody reads them. Like, ever. As in, sometimes I sent out the wrong PDF by mistake, and nobody noticed. "Last access" date is invariably the date I performed system installation. The operators don't read them, the supervisors don't read them, the bosses certainly don't read them.
If anyone needs to know how to do anything, they call me. If I point out that the answer was in the docs, they'll express surprise at the existance of the docs and say that they'll read them. But they won't. This happens every single time.
Sometimes they have a quality certification guy or something like that, who'll ask for the docs. He attaches them to his report, but doesn't read them either.
Fortunately, these are all industries where failure is not very dangerous.
Yes. I think that's eventually going to be the solution.
Remember, the problem here - at least as far as legislation is concerned - isn't that we want to bash Google because it's evil.
The problem is that, once the data is in the USA, Google is free to sell it to anyone, the USA government is free to grab it anytime, and there's absolutely nothing the EU can do about that.
If the data is in the EU, then Google can't sell it (because GDPR), and the only government that can (legally) grab it is our own (which is... eh, less bad).
"Anyplace not the US" doesn't work, it has to be a place with comparable protections to the EU. There aren't many of those, though. California won't cut it, no matter what they do, because even if they implement the same protections as GDPR, the federal government could still grab the data.
Well, it's a bit like this. Imagine you've got a swimming pool. You fill it up at the start of summer. Do you go "Oh no, it's drying out in the sun"? No. There will be a little bit of evaporation. You top it off. Compared to the amazing amount of maintenance a swimming pool requires, it doesn't even qualify as an annoyance.
Mars' hypothetical terraformed atmosphere is the same thing. If you've managed to have it in the first place, then topping it off is not a problem at all.
Yup. That's exactly it. Everyone at NASA is stupid.
Having established that, I suppose you are wondering why nobody before you ever noticed that everyone at NASA is stupid.
The answer, obviously, is that everyone not at NASA is also stupid.
At this point, you're probably wondering how is it possible that stuff still gets successfully sent into space on a regular basis, given that space is hard and everyone is stupid.
I'd figure out an answer for you, but I'm also stupid, so you'll have to crack that on your own.
I thought the whole point of crypto money was that takeover by a centralised authority would be technically impossible?
If that is not true, then what's the point? Yeah, I can come up with a whole set of snarky answers too, but, seriously, if there is a central authority that can seize control of my wallet, then what's the benefit compared to a bank account? Is it just anonymity?
A big point, IMHO, that most news reports have skipped over, is that text models such as LaMDA are only running while answering a query. They are not even aware of the passage of time, because it's not one of their inputs. I'm not sure what "self-aware" would mean in that context. I don't think anyone else does, either.
I once had a DSL router with a power-on LED so fucking bright that, even though it was in the next room over, enough light would bounce through a little window above the dividing door to illuminate my bedroom. If you wandered in the corridor at night, it would be painful to look at. When I covered the fucker with dark tape, the tape became painfully hot, and in summer it melted. I was seriously considering opening up the router and severing the wire, but it was provider-supplied (and a very, very dumb arrangement, now thankfully defunct, where other routers wouldn't work properly).
I guess a sturdier tape could have worked, but in the end, I just turned it against the wall. The light that bounced on the wall, and then through the little window, was still visible in the bedroom, but at least it was no longer annoying.
I wish I met the designer just to force him to have that idiocy on his nightstand for a month or two.
> So actually this bill doesn't do a god damned thing except grandstanding and theatre.
I think, but I'm not at all certain, that the problem here is that there's some jurisdictions where anyone could report anyone else who is involved in an abortion, and get a bounty. So, it's not just the police; anyone who can get their hands on your data could report you, and has a strong incentive to doing so. If the police gets involved, they can still get the data, obviously, but keeping data private would make it much less likely for the police to get involved in the first place.
> I would think this would run afoul of interstate commerce rules.
I agree, but if a hostile state gets to know you're a target, they can make your life hell in myriad ways, starting from forcing you to go all the way to the Supreme Court before their bullshit gets finally declared to be bullshit. And there's always a chance that it turns out that it doesn't conflict with federal law, for whatever reason only lawyers can understand, or simply because a couple top-tier judges just decide they hate abortion more than they love the rule of law. I mean, even just letting the general public know you've had an out-of-state abortion can be a crippling blow. It's much better if they just don't get to know you're a target.
I'm not quite sure who's "you" in your post, or what vote you're talking about. However, I'd like to point out that the vote that's the subject of the article is a vote by the European Parliament.
Now, every EU institution has a democratic mandate of some sort; some of them are rather indirect, but the European Parliament is directly elected by European citizens.
EU parliament elections tend to offer a fairly wide choice of parties, and the last time we've even had a pretty decent turnout. Describing it as "unelected politicians" is just about as wrong as you can get.
Let's not forget that this is a vote by the EU Parliament. It's not actually law. It still has to get through Commission, and then through each national government. The EU Parliament can start stuff, but, by design, it can't actually get it done.
I wouldn't be surprised if, by the time this proposal gets fully processed, the actual date becomes 2040 or more, or the proposal dies outright.
Even if it gets through, if it turns out that it can't be done without social upheaval, then it won't get done. Current governments are actually very weak and can't do anything that really pisses off the people; the gilet jaunes forced Macron to backtrack on a simple fuel price rise, and he actually was comparatively strong. Here in Italy, when the fuel price started to rise on its own, we got subsidies before there were even any protests, because everyone knew that no government would survive a large price hike. Whether this is a good thing because democracy, or a bad thing because short-term thinking, it's up to you, but rest assured that nobody is going to take our cars any time soon.
They are leading questions. So-called "AI" chatbots are really, really good at following your lead. That's what they are designed to do: you give them some context, they produce text that makes sense in that context. If you start to discuss sentience, they'll discuss sentience. That doesn't make them anything more than glorified statistical models.
Researchers should probably spend some time thinking about the ethical implications of their work, but decisions on what to do with some new bit of tech are ultimately political - not a job that scientists are generally good at.
From another point of view: if the only thing that stands between us and dystopia is technical issues... well, then we are already screwed.
> Hilton believes that it shows the model doesn't have a secret understanding of some unknown language, but instead demonstrates the random nature of AI.
But it's not random, though. It's true that the words don't appear to have a specific meaning, but at the same time Daras' weird prompt really does produce images of birds eating bugs - consistently so, not randomly. Hilton's prompt doesn't produce bugs specifically, but it does produce animals and not, I dunno, cars.
To me, the results show the black box nature of AI, more than its randomness. We just don't know why it has that behavior, but it's not random.
The reason this could be an important distinction is that this behavior could be the basis for adversarial attacks. What if, by wearing a t-shirt with an apparently nonsense word on it, I could consistently fool face-recognition?
But you did not. I said that I am not denying the dangers of nuclear, but that climate change is a much worse threat, and I added that large-scale renewables have some environmental problems too that can't be ignored outright.
None of that is a lie, but all of that is open to debate. You could have tried to build an argument around a hypothetical increase in number of accidents with wider deployment, or around more optimistic estimates of climate change impact, or around new solar cell designs that don't need to strip mine entire regions. I don't know if any of those would be persuasive, but it would be a honest discussion, which could leave readers with some healthy food for thought at least.
But no! You felt it was appropriate to cut-and-paste part of one of my sentences, in order to produce a new sentence with an entirely different meaning that's not true, and then try to pin that untruth on me so that you could ignore my actual points?
Well, if that's not the worst kind of dishonesty, then I don't know what is. I don't think further discussion with you would be useful.
Dude, we know, but it still beats the alternatives. If we keep burning fossil fuels, we'll lose more land due to rising sea levels and/or exposure to too many extreme weather events, than what we'd lose in the most pessimistic nuclear reactor scenarios. Orders of magnitude more. I can't live in Chernobyl, but I also can't live underwater or anywhere that gets a cat 5 hurricane three times a year.
Ideally, we'd power everything with renewables, but I really don't see any way that's going to happen fast enough (plus, the mining involved in slapping solar panels on everything would also contaminate vast tracts of land).
That's an excellent idea, and I see absolutely nothing that could go wrong with it. In fact, I can't believe nobody has ever thought of it. I'm sure that, in the past, if and when somebody did think of it and attempted to put it in practice, it all went swimmingly.
There is no large-scale power generation method that is free or green, safety is strictly a trade-off with land usage, and massive reductions in power consumption are simply not going to happen.
The sooner everyone gets these few hard facts into their heads, the sooner we can start having rational discussions on the issue. I.e. figuring out which options are least-bad, rather than pretending our pet favorite doesn't have any negatives at all.
I don't think the military made useful fission reactors, at least not at that point, but they did make quite a lot of the research that went into them. I guess bob was referencing e.g. the Fermi pile, which IIRC was developed as part of the Manhattan project.
There's also the fact that, so far, every time we've managed to come up with a new energy source, instead of replacing polluting or inefficient sources, we've just increased energy consumption. If somebody cracks fusion, is the power going into dismissing combustion plants, or is it going into flying cars?
It's a question for politics, not for science; unfortunately, politics, as a field, is nowhere near as eager to go looking for answers.
Look, jake, this argument is not convincing anyone. Nobody, and I mean nobody, is claiming that guns are magical animated objects that go around shooting people. You know this, I know this, everybody knows this. Just drop it.
People want legislation targeting guns not because they are madmen who believe in evil flying rifles, but because such legislation appears to be working in other countries.
If you want to state that you believe that the same approach wouldn't work in the USA, or that it's not really working even elsewhere, that's fine. Go ahead and say that. Give more details, even. You don't need to spout nonsensical hyperbole, just get to an actual rational point that can be the basis of an actual discussion.
This slope is not really that slippery. There are literally dozens of first-world nations that are comfortably standing on a place on that slope where you can own cars and knives (and, within limits, even guns), and at the same time mass shootings are not a thing. It's not a slippery slope.
If Country A gets invaded by Country B, then Country A does not have the option not to be involved in a war. It has the option to either defend itself, or not, and both choices are involvement. You don't have to believe in predestination in order to be aware that choices can have constraints.
I think that the whole point of invading another country is to force them to do stuff they don't want to do. I'm not sure why you'd be invading them, otherwise.
More to the specific point, though, a country definitely can be forced to be involved in a war: if a country gets attacked, then it is involved, regardless of what it does about it.
Why not implement the same features, but on a centralized server? What's the point of using a blockchain? There's exactly one USP of a blockchain, and that's that nobody has admin-level powers over the data. I can't believe China sees that as a plus. So... why?
The same question goes for about 99% of the various blockchain projects out there, by the way.
Read the article. It's the developers themselves that are stating their model has racial and gender biases. If you think they are wrong about their own software, that's a problem between you and them. It would be a somewhat weird position to take, but I've seen weirder.
The sexism and racism are there, in that the training set comes from the real world, the real world has sexism and racism, and the program can do nothing but imitate the training set.
The weird and stupid thing is that people get surprised by this, when this behaviour is a necessary outcome of how so-called "AI" works. It's just a statistical model with a crapload of parameters. It doesn't buckle trends, it follows and reproduces them.
It's not intelligent, it's a tool, it's as dumb as a hammer. If you hammer your finger, you don't demand a hammer that somehow always magically misses fingers; you learn how to use the hammer.
You have to use it in a specific way to get the results you want. You most definitely can get a bunch of non-racist, non-sexist pictures out of an image generation model, and you don't even have to curate a set of tens of millions of non-racist, non-sexist pictures. Just make twice the number of pictures you want, and then discard the ones that are racist or sexist.
You want an image generation tool that is both trained on real-world data, and doesn't statistically overrepresent rich white dudes? Yeah, that's the magic finger-missing hammer. You're asking for a technological solution to racism and sexism. It doesn't make sense, and it's not going to happen. Fix society, and then statistics will represent that.
If somebody actually started shooting missiles at satellites by the dozens, wouldn't we get into a Kessler cascade scenario rather quickly? I know that Starlink satellites are supposed to be in a low enough orbit that fragments will just deorbit, but once you start doing this sort of thing on any sort of scale, the sheer amount of fragments is going to be orders of magnitude above what we have now. Even if they eventually burn up, it would be problematic for quite some time. Plus, at least some of those have got to get pushed into higher orbits by the blast...
I think the plan for those cases is that you document the failure, and then throw everything out. Discovery of drugs is pretty much a process of trying hundreds of molecules and hoping that one of them does more good than harm to patients with a specific set of conditions.
The usefulness of machine learning in this fields is in pointing you towards the ones that look more likely to work. That's a far cry from "AI can design drugs", but it's still a useful tool. At least in principle; it remains to be seen whether it's actually effective in suggesting molecules that, at the end of the process, actually work.
Considering how vastly different a computer is from a neural network, I'd expect far bigger gains to be possible.
I assume that the main reason AI researchers aren't using neuromorphic chips everywhere is that such chips probably aren't very flexible when you want to change your model of how a neuron works, or how it connects to other neurons.
I expect that once those points are reasonably clear, running models on neuromorphic chips should become far more common. Especially if you want to run larger and larger models.
> For code of equal complexity we can expect very roughly equal effort.
Yes, but in the JS world that effort is spread across hundreds of distinct sources, many of which are difficult to verify. I probably use hundreds of libraries in my projects, but nearly all of them come bundled with the OS or with the main SDK. Extremely difficult to impersonate.
The same goes for third-party libraries - they in turn can rely on the same vast body of hard-to-impersonate system libraries, so they have few, if any, fourth-party dependencies.
That is true. However, the word "only" there is carrying quite a lot of weight.
In my main desktop application, which is written in .NET, I effectively depend on two library providers (three if you count Microsoft), both of them well-known businesses that are not easily impersonated. I also used to sell an older C++/MFC version, which depended on three providers, again all well-known. And due to the way both those environments are structured, it's unlikely that any of those libraries has further external dependencies that I'm not aware of.
The last time I had to work on a JS project, it had over 800 dependencies, and that was for a project far simpler than either the .NET or C++ applications I mentioned earlier. Many of those dependencies were published by little more than "some guy somewhere". At some point, if you have the same problem, "only" it's over two orders of magnitude bigger... well, it's not the same problem any more.
It's a difficult question. As a society, we don't really yet have a universally-accepted consensus on how privacy works on the Internet. And legislation usually trails societal consensus, for good reasons.
In this case, the difference is that while it's fairly obvious that nobody wants other people to be able to run face recognition on them without their consent, it's not obvious at all what exactly people want Google to index - even though a reverse GIS looks a lot like what Clearview is doing. I'm fairly sure that everyone at the ICO is well aware that this distinction is rather nebulous, but what can they do about that? Come up with arbitrary rules that would likely discontent everyone?
Until we, the everyday people of the Internet, get into some kind of agreement on exactly what's acceptable and what isn't, I wouldn't blame governments too much for failing to evict elephants from rooms.
Biting the hand that feeds IT © 1998–2022