Sort of. The secret someone in your example goes to a bunch of effort to make the one ball not in S a green one while all the other balls are red. You're colour blind so you have to sample the balls as you say. The QC is not color blind, and has been told in advance it's looking for the green one, so it can just grab it. Except it gets it wrong most of the time so you actually have to do some sampling anyway.

quantum computers are not just faster calculators, they sometimes need far less data to do a job.

No, the QC isn't a faster calculator (typically it's much slower) and it doesn't need less data (usually it needs much more). It can, however, perform specific types of matrix multiplications very quickly, so if you can contrive your problem to take advantage of that the computation might be faster. It depends a lot on how much contriving is necessary. It's much the same as how a big pipe with a fan on one end is really good at doing fluid dynamics computations.

There's no magic. It is literally just simple matrix multiplications.

The OMG quantum is weird crowd like to talk about entanglement as if it's some strange situation. Rather, massive entanglement is the norm. The strange situation is when some monkeys with lasers come along and very carefully disentangle a couple of electrons or whatever from everything else in the universe then let them just barely become entangled with each other.

The GP used "observe" but it's entirely unnecessary.

The reason the world doesn't do any of the "weird" quantum stuff IS that it's massively entangled. We can only get the weird stuff by setting up very delicate, very special cases. Then, when we "observe" them, i.e. entangle them with a giant measurement device of some kind that is itself massively entangled with the rest of the universe, the special delicate state "collapses" into an ordinary massively entnagled one.

So surprising.

Fortunately he has a nice shiny law to use, which the summary mentions he already threatened.

The other possibillity in his threat, listing Anthropic as a supply chain risk, means that any company with Anthropic anywhere in their supply chain is potentially banned from government contracts. In the US that's pretty much the corporate death penalty.

He'll go with direct force though.

There's no such thing as "instinctive equilibrium." All organisms'* populations increase until they reach their environment's carrying capacity. That's the point at which their death rate equals their birth rate. The feedback isn't instant so usually the population oscillates around the actual carrying capacity, i.e. feast followed by famine.

* Except one, maybe. Humans did just as all other organisms do up until recently. It's called the Malthusian trap. Every increase in food production capacity was just eaten up by increased population, then right back to famine and disease. Recently, fairly developed countries (not "the west") seem to escape the trap, primarily by educating women.

I think Anthropic named two for you. Okay, the second one isn't a national security threat, it's a desire for a tool to deal with them.

The process names haven't actually indicated feature size since about 1997 (~250 nm). It's supposed to indicate the "equivalent" of a 90's era transistor. Leading edge EUV machines have a resolution of about 10-15 nm, and most actual features are considerably bigger than that.

You CAN push around individual atoms, and you can do some really cool stuff that way. It's impractically expensive though, and impractically slow to make anything that's not a tech demo.

Yes. Efficiency depends on both what the numerator and denominator are. Your boss doesn't care about your other outputs. You do.

Yes, and the human is probably doing the work at a computer anyway, using a significant fraction of the electrical power it would be using if it were running a model. Probably more electricity overall because the human takes much longer to complete the task.

The point is that Sam Altman is not wrong... on this specific point. AI in general is quite efficient. Paradigm shifting efficient in lots of cases. The wild energy needs Altman keeps talking about are because he wants to be the first to make AGI and he can't wait for algorithmic improvements so he just wants to throw unprecedented amounts of compute at it.

Yes, the brain is doing a lot more than the specific task. It's probably true that the brain is more efficient on a per operation basis. It's generally less efficient on a per task basis. The cynic would point out that the off-task stuff your brain is doing is a detriment to your employer, not a benefit.

You, like most people, including Sam Altman, are stuck on LLMs. ANNs can do a lot of things and they tend to be vastly more efficient than humans and quite a lot more efficient than conventional algorithms. I used to ask labs what dream projects they would do if they had thousands of grad students each with infinite patience; that's the kind of thing that's an ideal application for AI.

That's also the EU, so it excludes the UK and Sweden.

Price is really a terrible way to compare military strength. Finland's entire military budget is less than $7 billion USD a year for 280k active members and 900k reserves. That's about what the US is spending on upgrading Guam's missile defences, or about half what it's costing to station 5000 marines there.

There's a startup in the US that apparently reaised $100 million for particle accelerator-based photolithography, but they're getting quite a bit of skepticism. There's also university research into new technologies of course, but I don't know of any really serious push or anything likely to deliver "soonish." It took ASML thirty years to get EUV working.

The Chinese are certainly investing heavily in lithography for obvious reasons. They might have something in the ballpark "soonish" or they might not. I wouldn't hold my breath on equivalent or surpassing though.

A standard Big Mac is ~570 kcal, or 2.4 x 10^6 J, about 95% of which can be extracted by the human digestive system. An H100 uses about 350 to 700 W max, with typical working draw on the lower end. Say 500 W, being generous.

A Big Mac would power that H100 for an hour and a quarter or so, which is a lot of queries. You could get a typical diffusion model to make a few thousand pictures in that amount of time while a typical artist might have difficulty finishing one on a single hamburger.

For many things AI inference is quite a bit more efficient than a human and it is generally much more efficient than a conventional algorithm where comparable conventional algorithms exist.

But ultimately it's an academic question. Capitalism did evolve in a particular way, and it's very different today from what it was 50 years ago.

True. It's harnessed by effective taxation, sophisticated financial and insurance systems with effective safeguards, strong social programs treasured by citizens and widely shared prosperity that lets pretty much anybody own and utilize effective capital if they desire. Even traditionally disadvantaged parts of the world are vaulting ahead with their own social and economic innovations and enjoying unprecedented prosperity.

And then there's the United States.

It's also comical when you look up the numbers and learn that Europe has something like twice as many troops as the US, more tanks, more artillery, more of pretty much everything that's relevant to actually defending Europe.

The idea that Europe is weak is propaganda or, at best, based on a naive comparison of spending. The US military is extremely expensive because it's based around the ability to project offensive power anywhere in the world on short notice, including in places where there's no friendly support.