Exactly. And Vmware's calculus is that as long as it has a dozen gigantic clients (governments, huge enterprises) that it can charge whatever it wants, then the small fish aren't worth it.

The target audience for this particular rule of the AI Act isn't the producers or users of AI. It is the consumers of content that might be AI generated.

And frankly, letting people know if they see AI content or not is a fair thing to ask.

Fun fact: The EU AI Act considers this. It makes a difference between "AI generated" and "AI modified".

But it's hard to put this into the type of watermarking they are using. Because if you generate a text and then change a few things here and there what you are doing is degrading the watermark. But since the watermark is probabilistic, it's hard to tell if it was a weak watermark from the start or if it was degraded.

That ain't the same thing.

This thing is doing things you didn't tell it to. It might even do things you explicitly told it to not do. It is not a case of "well, you could've forseen this as a consequence if you had been careful" like your campfire. My example is only a bit silly, it is absolutely possible that an AI agent will decide to hack NASA because you asked it for some data for a school project. Even in the broadest possible sense, that does not fall into stuff that the law so far has considered to be in the domain of negliegence.

AI agents do stuff you could not possibly have expected. Heck, I asked Claude a fairly simple knowledge question yesterday and it wanted to read from /dev/null. There is no argument you can make that would convince a jury that I should in any way have expected that and taken precautions. In a hypothetical world where a read from /dev/null randomly kills a puppy somewhere on the planet, I don't think I would've been convicted.

So basically create a permanent underclass of indentured servants...

We have bubonic plague infections every year in the United States. Very, very few, admittedly, but it's not like it's been eradicated. So there's still benefit in studying it.

The more I hear about this so-called "SF Democratic Party," the more I'm convinced they don't speak for the majority of SF voters. Seriously. They seem to be some strange, centrist organization that no voters belong to and nobody knows where they get their money from.

I don't think people looking for entertainment are going to pay to watch a propaganda hit piece.

The question was about sea level rise.

If the rising waters coincide with other climatic events, such as rainstorms and high tides, the phenomenon could swallow beaches, chew away bluffs and coastlines

There is a narrative that oh, a few cm of sea level rise is no problem, it's just a little higher water. It isn't.

However, the kinds of multi-million-dollar mansions that celebrities buy aren't the rickety perched-on-the-edge-of-a-crumbling bluff houses built in the 30s and 40s. They tend to be built-by-high-priced-architects homes that had civil engineers to advise about the stability of the foundation.

https://www.cbc.ca/news/entert...

If AI is ruled a serious risk that requires a government kill switch you won't be running any open weights models on hardware you control. I expect that's what the FSF is afraid of. That's why they refer to it as hypothetical technology as in "assuming such a kill switch was developed".

It might seem far fetched but consider that you are probably (based on the American readership of this site):

- not permitted to buy more than a few grams of effective cold medication per day and have to show ID, which goes in a tracking system, in order to do so.

- while you are allowed to purchase microcontrollers, if you install this on them they are illegal to possess

- possession or use of all manner of chemicals, combinations of chemicals and shaped metal considered dangerous will get you prosecuted with various degrees of zeal

The big AI companies, who already possess the most effective possible actual kill switches, seem oddly enthusiastic about the whole thing.

Not just downlink, but the inter-satellite bandwidth and latency is also likely to be a problem. I saw an analysis of what it would take to build a big datacentre, and there the cooling really does become a problem. Not the radiators so much as the plumbing to move the heat around. So everyone is talking about lots of little satellites, but that means you're restricted to a handful of cards locally. It's going to restrict the kinds of jobs you can do when the next rack over is a thousand kilometres away.

That's how it already works.

It's not. Full disk access is sticky. When you grant it, the app has it until you go into settings and revoke it. It's not like sudo, it's like running that app as root, with no permission required after the first time. On Linux to get the same effect you have to edit a text file with root permissions and give the program passwordless sudo privileges, which should make any competent Linux users' hair stand on end.

And the next step will be "See, we can't allow full disk access." And then macOS will cease to be usable.

People have been "predicting" this slippery slope thing for ages and it hasn't happened.

These satellites would be in LEO

Sort of. They'd be in probably lowish sun synchronous dawn-dusk orbits, a narrow band following the terminator where the sun always shines.

and fall back to Earth after 5 years anyway.

Yeah. Thus "you generally need to rebuild your entire infrastructure... every five years or so."

The server hardware will be obsolete by then anyway.

The server hardware maybe. Not the power plant, (extensive) cooling hardware, infrastructure. These aren't GPUs floating in space, they're the entire computing, communications, cooling, power etc. infrastructure. The every five years thing would be like building a datacentre and everything supporting it in the middle of the desert and then nuking it every five years and starting over from absolute scratch.

If the cost to orbit is low enough, and there are no land taxes or electricity costs in space, and it solves the locality and latency issues for them, it starts to make more sense.

The cost to orbit, whatever it is, is pretty much money on top of whatever it would cost to build in a remote location on the surface. Taxes are minimal, zero or sometimes effectively negative for remote development on Earth. If you want to build a datacentre in the middle of the Sahara you can probably convince Morocco or someone to give you money. Solar panels are bit more efficient in space, and you get to use the 24 hours a day, but you also have to toss them every five years and buy (and launch) new ones. Along with everything else. And latency is going to be worse, not better. Your servers are whipping around the planet in polar orbits on the terminator. The minimum distance they are from any point on the surface is going to be about 300 km and that's going to occur for a few minutes a couple times a day. Most of the time they're going to be half the planet away.

I watched Scott Manley's video when it came out. He's right, the technical issues everyone seems to focus on aren't great, but they really aren't that bad. The economic and latency issues are the killers.

As far as I can tell nobody really has any consistent idea what "kill switch" is supposed to mean. Eg. https://www.scientificamerican...

OpenAI and Anthropic already have kill switches, from the power cord through to account bans. If they don't want anyone using their model they can do that. Likewise, if the US government is sufficiently motivated to order a service provider to stop providing that service, they can absolutely do that. I get the feeling the actual companies are hoping they can implement something that has lots of buzzwords and does sort of what their "guard rails" do, and everyone will stop bugging them about it.

Naturally the FSF imagines some hypothetical technology that allows the government to reach into your home to shut off the matrix multiplies running on your own computer.

Excellent pedantism sir!