ISTR the FDA having some sort of authority over pointing lasers at things you don't own. Might be illegal. A focused LED source might pass, not sure there are any general laws prohibiting nuisance photons so long as they're non-coherent.

Guess they took the whole "given enough eyeballs" thing a little too seriously.

Me neither, not sure what movie GP was thinking of, but in the interest of going completely off the topic: I went in a post office the other day and the fire alarm panel was beeping. Went over and looked at it, as one does. It said "H2 alarm, battery room." The counter staff didn't seem to be concerned. I did my business and left promptly. Building was still there last I looked...

Back to the point, if we're going to have data centers, I guess we could do worse than powering them with nukes. It'd be rather nice if we could get them to usage-shift, take up the extra capacity at night when demand falls. Perhaps once the lizards implement the One World Government, we can coordinate that globally.

Having read most if not all of the Robot novels over many years, as I recall, the usual construction is they spend a paragraph introducing the Three Laws, then the remaining 300 pages explaining how they fail, backfire, or get exploited in order to produce the central conflict of the story.

I do that all the time. Quite often the most effective way to get a human on the line. Don't like my language? OK, then give me a 'press zero for a human' option.

I'd be interested to see the power budget on this. Regular RAM's power cost mostly scales with installed capacity, compressed RAM's power cost is going to scale with usage. It might be a big win for specific applications that need to have huge amounts of fast memory on-hand but aren't hammering it constantly. Doesn't look like Meta published any power figures.

If the compression power cost is low enough, it might be a win for everyone. Otherwise, it's still a win for the hyperscalers: they get to exchange physical bought-and-paid-for RAM for increased electrical usage, which is partially subsidized by everyone living in the area (and everyone in general, if environmental effects are considered.)

To your point, it looks like something the average desktop or even server operator isn't going to use. Certainly not small embedded systems -- this approach requires specific compression hardware.

I loved Art Bell. He'd just tell the perpetual motion machine loons "you can't do that, it violates the laws of thermodynamics, show me or GTFO." George Noory, the last thing I remember about listening to his show, he had the head of some medical examiner's office on, and conducted a normal interview, and then at the end, "have you ever seen anything that looked like an attack by a vampire or werewolf?" You could hear the guest rolling her eyes, and she said "we... Don't see a lot of that."

Quite. Invoking a self-destruct mode within sight of whatever cop wants the data is a Bad Idea. There's room to argue whether this is how it ought to be, but this is currently how it is. Don't wipe your phone after being asked to hand it over, unless you KNOW it contains data that will get you (or others you care about) in more trouble than a 'destruction of evidence' charge, or unless you want to make an expensive stand on principle. There are legitimate folks who live in that gray area: political dissidents under hostile regimes, some whistleblowers and journalists, abortion patients in societies where it's "health care" on one side of an arbitrary line and "murder" on the other, etc.

Might get away with it if the destruction doesn't follow from a specific action taken after access has been demanded, such as:
1) configuring a 'wipe after n bad passwords' setting
2) setting a duress-wipe code and writing it on a post-it, but not entering it, hoping the cops try it
3) setting a duress code/fingerprint that reboots but doesn't wipe -- that makes recovery much harder, but under the current framework, this might lead to a legal battle over whether crypto keys in RAM are "evidence" subject to "destruction."

I'd rather enjoy reading the court proceedings over any of those options, but I'd not enjoy it enough to be party to said proceedings.

GrapheneOS has an 'idle reboot' function like the iOS feature at issue here, and it's unknown whether there are any similar vulnerabilities in that. People should probably turn it on, it's more likely to protect than to harm.

I think probably the best advice is, for those worried about non-criminal personal data while traveling, take a travel phone and access your data over a VPN when it's safe. For criminals, maybe leave your phone home while doing crimes and don't store evidence on it.

But what about Taco Bell?

Or perhaps it's the plot of the latest direct-to-streaming Terminator movie, in which the clankers are sent back through time to dress up as plaintiffs and use their prodigious legal-argument skills to thwart this attempt against Skynet!

... Back at base, bugs in the software
Flash the message, "Something's out there" ...

Someone told me once that the hardware in there is this mediocre Android dev board with a little additional panopticon gear bolted onto it. So yeah, definitely in the "Free phones handed out in front of Walgreens" territory.

Not sure what would be the utility of running that sort of software on an individual endpoint machine, there's already much better malware intended to run on those. But it does point out the ease with which a sufficiently motivated actor can roll out an impressive surveillance grid with cheap (or was cheap before the AI bubble) off-the-shelf hardware.

Did we really expect more from the company that also left a lot of its devices searching for WiFi APs with predictable SSIDs and passwords, then offering an ADB port? Or, for those that weren't actively connectable, requiring pushing the exposed power button on the back with a stick to make them connectable?

Startup culture, move fast, break things -- preferably someone else's things.

Probably some of that, but it's also pretty hard to prosecute someone for a specific act absent evidence that they intended that act to occur. There's a whole family of "negligent X" crimes, but those tend to target specific outcomes ("that person is injured/dead because you were negligent"), and I don't think there are any negligent hacking statutes on the books.

Doing a hack manually is clearly prosecutable. Writing and invoking a script to do the same hack automatically is clearly prosecutable. Specifically directing an AI agent to perform an action that the person giving the command knows to be criminal is probably prosecutable under innocent-instrumentality doctrine. Telling the agent "get me a copy of the latest Toy Story" leading to the agent hacking Pixar and downloading pre-release video, less so.

I think the closest analogy in these cases would be either vicious-dog prosecutions ("you knew your dog was dangerous, you frequently brag in public about how vicious your dog is, you failed to contain it properly, it injured someone") or corporate-officer responsibility ("As CEO, you may not have issued the order to store the food in the rat-infested warehouse, but you had a duty to ensure that didn't happen, and you knowingly allowed it to happen.")

Certainly can't prosecute the model for hacking Acme when it was told to pen-test EvilCorp, and under current law, I don't think we can prosecute the creator or the person who gave the "Get me EvilCorp's R&D files by any means necessary" order for the actual hacking of Acme. And I think the vicarious-liability laws on the books are deliberately pretty narrow, to avoid them being used for all sorts of BS prosecutions. Might be able to get a vicious-dog conviction for sandboxing the known-dangerous agent by removing the default route and hoping it doesn't notice.

We'll probably need new laws for this. For now, if anything, it'll probably be civil litigation along the lines of negligent supervision -- given that the big AI companies are currently privately held, civil penalties might actually even hit almost the right people. Once they're public, it'll be pension funds and 401k's bearing the penalties for their corporate malfeasance, at least criminal law catches up with the concept of non-conscious agents that are capable of forming functional intentions, possessing "knowledge" of which targets are permitted and which not, and selecting actions against out-of-scope targets anyway.

One of those logging trucks with the giant claw?