Cooperation would, indeed, be better. Unfortunately both sides want to ensure that only they win.

China is the enemy because Realpolitik asserts that there can be only one top country.

I don't know. Having people working there under assumed names sounds sort of like "Manhattan project" level security. I.e. it's a big secretive project with a lot of government backing". I'll grant it's a wildly overused metaphor, but that just means everyone will recognize it.

Sorry, but copying from what others are doing is very normal. Copyrights and patents were originally intended to be quite temporary in duration. And that's as it should be.

I'm all in favor of temporary copyrights and patents, say 5 years. Perhaps 10 if there are a LOT of up front development costs. Beyond that is an aberration, and one shouldn't expect others to abide by it. (And the US basically ignored UK patents and copyrights until quite recently.)

This is a lab machine, and it's not clear that it's making large chips. I think your 5-10 year prediction of last year is probably right. There will be engineering challenges in converting a lab machine into a production machine.

Actually, my (uninformed) prediction last year, and this year, is that it will take about a decade for China to equal the production of TSMC assuming TSMC keeps improving. But that they'll have "good enough for 90% of the market" within a very few years (and perhaps already do).

It's not that simple. Every holder of power acts to restrain challengers. If you allow monopolies, then innovation in that area slows drastically. When you have diverse centers of development, then development tends to be faster...but more expensive.

So if you want the most profitable companies, then you allow monopolies. If you want the fastest development, then you break up monopolies, of prevent them from ever arising...but this will make the companies less profitable (on the average).

Historically democracies have been more willing to break up monopolies. Right now, though, the US doesn't seem to be willing to do so. So now rapid development depends on competition between either countries or blocks of countries.

I don't know what the History Channel said, but Germany was many years away from making the atomic bomb when the Nazi's went on the path of expelling the intellectuals. They had most of the theory, but so did everyone else. They had the people who could have helped convert the theory to practice, but they expelled them. But this was multiple years before theory was converted into practice (i.e. "The Italian navigator has landed in the new world. The natives are friendly.") At at THAT time, the US government didn't really believe in atomic power.

I specified "single threaded", which is true for most of the code I write, even that used by multi-threaded routines. (It means that a lot of the reference parameters need to be const, but that's minor.)

FWIW, I find even C++ to be annoyingly overprotective in the wrong places. It causes me to need to write multiple copies of the same routine that differ (nearly) only in the parameter specs. E.g. when the looser version would be safe anywhere, but can only be used by routines within the class.

I'm inclined to agree, but thinking about it there might be some things that an "agentic" AI could help with. Like "fill out my timecard for today" or "every time Outlook web logs me out, log back in with my credentials." You know, the things that would give the bureaucrats a heart attack if they knew I could do them instead of wasting my time.

Assuming I trusted the AI enough, of course.

And in a single threaded loop it feels like unnecessarily having one hand tied behind your back.

To be fair, C practically insists that you use raw pointers. I think the C standard should allow references. Also some way to handle unique_pointer and shared_pointer. (I mean a way that's standard for the language.) But this would require that the pointer know how large a chunk of memory it was pointing at.

More specifically it's worse because you can't have multiple references to the same memory location even within a single threaded piece of code. It's like uselessly having one hand tied behind your back.

That would, indeed, probably be a good solution. The doing of it, though, is "not simple".

If you could trust an LLM not to hallucinate, that would be a good job for LLMs. There was a system called PLATO that tried something like that several decades ago, but it was both much too expensive and much too limited. Also much too inflexible.