Assuming all goes well, I agree. OTOH, returning old equipment back into service after it had been written off is a little risky, don't you think?

It's perfectly possible to build a machine that will incapacitate a human at close range; all of the world's militaries have been doing it for years.

It's also possible to deliberately dial down the machine's combat effectiveness to the point where a fight against a human might not be a foregone conclusion.

OTOH, having done the latter, what have we learned? That we can design and build a robot that we nerfed just enough that it usually loses in hand-to-hand combat?

As they say, what's remarkable isn't that the dog plays a good game of poker; it's that the dog is able to play poker at all.

As methods for by suicide-by-wealth go, it's classier than drugs, and more accessible than a carbon-fiber submarine. I predict we'll see some pretty spectacular Youtube videos of people making themselves (and their car) two-dimensional.

I wonder how the performance of 2 networked 512GB Mac Studios would compare to the performance of a (hypothetical) single Mac Studio with 1TB of RAM?

My intuition says that the network linking the two machines would be a bottleneck, but I don't know if that's a minor performance problem or a major one.

I didn't know either, but I do know what JPEG is, and I have a pretty good idea about what XL might mean, so I was able to suss out that it might be a new and improved lossy-compressed image format.

Of course, approaching the problem rationally like that doesn't provide me an opportunity to make childish insults, but that's just one of the downsides of being a grown-up that I'll have to live with.

The real killer app for AI will be when someone creates an AI that grovels over all the millions of low-effort AI slop games out there and flags just the interesting/fun/novel ones for your attention. Otherwise you'll never find the gems amongst all the dross.

There's no reason why Python (compiled from Rust) would be any slower than Python (compiled from C). Both languages compile to native executable code, and both have good optimizers available.

There's nothing wrong with knowing your own limitations and acting accordingly. Had Mr. Zuckerberg had that sort of self-knowledge, he might have avoided lighting 80-billion-with-a-b dollars on fire and renaming his social media company after a technology that nobody wanted.

That is true, but it's a little more complicated than that.

The average driver is pretty good, 99.9% of the time. The other 0.1% of the time, they are tired, or angry, or momentarily distracted, or just looking the other direction from where they need to be looking under those particular circumstances. It's those occasional blips of inattention that cause lead to most accidents occurring.

Yes, I agree. The problem with thinking "this bug seems harmless, because I can't imagine how anyone could exploit it" is in the "I can't imagine" part; my imagination is limited to what is covered by my mental model of how computers work, but an attackers' ingenuity is not.

In particular, the C/C++ optimizer is a devious beast, and will exploit any opportunity to make the code more efficient, even if that means doing things that are wildly unintuitive to a naive human reader -- and it sees any instance of undefined behavior as an opportunity to exploit.

(C) is the only one of those the AI companies are really concerned about -- they want to stave off model collapse for as long as possible. The problem is that AI generates so much content that the ratio of AI-generated data to human-generated data keeps rising, and sooner or later there simply won't be much human-generated data left to anything to consume. At that point they'll either have to figure out to ingest AI-generated content without causing degradation, or they'll have to accept that AIs have gotten as smart as they will ever get, and it's all downhill from there.

It's really the same problem that we see with quantum physics -- the subject matter is esoteric enough that common sense can't be used to determine whether what you're listening to is really advanced theory or utter BS (or some combination of the two); they both sound the same, and the situation isn't helped by the presence of large numbers of people who think they know what they are talking about but don't. :/

I think the distinction they are making is between technology that kills people, and technology that entertains people. If someone fitted actual weapons to a Doom engine, they'd probably object.

Let's make the hopeful assumption that the superintelligent AIs will not "go rogue" in any way, but will correctly try to carry out their tasks as assigned by the people who designed and trained them.

Pretty much every major nation has military AIs, at this point. So the Chinese AIs are working hard on how to defeat the Americans, and the American AIs are working hard on how to defeat the Russians, and so on. In the final analysis, no matter where you live, there is probably an AI that has been explicitly tasked by its operators with figuring out the most effective ways to do you in. Not because they've "gone wrong", but because they're working as intended.