achierius

Generally yes. Or yes, you could just do it yourself in your free time.

> It hasn't happened yet.

Not to you. But it has happened to tens, hundreds of thousands already. Did you miss the whole 2016 election news-cycle?

Then why bother? Why not go all the way, try to find a way to a new, better system, rather than gambling that these people who so totally hate you would one day become willing to compromise in order to save the current one (with benefits then most of all, not you)?

That's assuming these benchmarks are the best we're ever going to get, which they clearly aren't. There's a lot to improve even without radical changes to how things are done.

I can at least give my favorite and least ...

- most enjoyable: fiddling with new optimizations

- least enjoyable: root-causing bugs from customer crash stacks

There's more than that. Not a huge amount more but still.

- multiple large companies contribute to each of the larger AoT compiler projects; think AMD's contributions to LLVM and GCC, and multiple have their own internal team for handling compiler work based on some OSS upstream (eg Apple clang)

- various companies have their own DSLs, eg meta's Hack, the python spinoff Goldman has going on, etc.

- DBs have query language engineers which are effectively compiler roles

- (most significantly) most hardware companies and accelerators need people to write toolchains; Triton, Pytorch, JAX, NVCC, etc. all have a lot of people working on them

It's definitely a pretty small world, and to make things worse there are sub-niches -- between which there's certainly cross-pollination, but that's still a barrier to people looking to change jobs. Frontend language semantics (where most PL papers focus) vs. middle-and-back end optimization and hardware support; AoT compilers vs. JITs; CPU targets vs. a blossoming array of accelerators, etc.

Beyond that, I've definitely interviewed people who seemed like they could have been smart + capable but who couldn't cut it when it came to systems programming questions. Even senior developers often struggle with things like memory layouts and hardware behavior.

Have you tried having an LLM write significant amounts of, say, F#? Real language, lots of documentation, definitely in the pre-training corpus, but I've never had much luck with even mid sized problems in languages like it -- ones where today's models absolutely wipe the floor in JavaScript or Python.

All the ones I've had, and most of the ones I've seen, we for bespoke compilers and toolchains for new HW / specific languages

Social dislocation, a loneliness epidemic, the breakdown of civil society and of trust in the media, gambling unleashed in everyone's pockets, billions of dollars spent trying to get you addicted to scrolling on your phone... Yes, it's worth bemoaning that.

Why should you "control" for these variables? AIs will effectively be punished for doing various inscrutable things by their own internal preferences.

Most compilers do not use "unit" tests per se. Much more common are integration tests targeted at a particular lowering phase or optimization pass.

This is pretty important since "unit tests" would be far too constraining for reasonable modifications to the compiler, e.g. adding a new pass could change the actual output code without modifying the semantics.

What else comes to mind in terms of inefficiencies? I can think of quite a few but you seem to have deeper insight as to their ranking so I'm curious as to your thoughts.

They also have a Micron trademark for software, as they sell some software packages. So it's really just a question of whether or not they notice and decide to go after you.

Do you not think the same thing would apply in the US? If we were having serious domestic conflict there would be veterans on both sides, not to mention that other countries would certainly train officers to help their preferred side. The real point is that "technical superiority" (air power, artillery, mechanized equipment) is not on its own sufficient to win a war when the populace is opposed to the occupying military presence, and that military is not willing to totally butcher that populace.

Great, but what about the people who still starve to death? The ones who are left to die on the streets? Yes, those homeless now have phones, but I imagine that does little to mask the hunger pains. And unlike housing, food has not gotten significantly more expensive -- so what gives? The fact that we let this happen, even in the very heart of the richest nation on Earth, the richest nation in history, is proof positive that regardless of how rich the world gets, there's no guarantee that those riches will be used equitably. I agree that we've made massive strides; as such, making more massive strides is unlikely to change this dynamic.

But is that the address being returned by mmap? Furthermore, what instruction is this crashing on? You should be able to look up the specific alignment requirements of that instruction to verify.

> ChatGPT and Claude and Grok all agreed one way or another, with various prompts.

This means less than you'd think: they're all trained on a similar corpus, and Grok in particular is probably at least partially distilled from Claude. So they tend to come to similar conclusions given similar data.

No, this is definitely a thing in certain crowds. Not everyone, definitely not everywhere, but some circles really love this stuff.

You have it backwards. Modern compilers don't use SSA because it's "simpler", we use it because it enables very fast data-flow optimizations (constant prop, CSE, register allocation, etc.) that would otherwise require a lot of state. It doesn't "pretend state doesn't exist", it's actually exactly what makes it possible/practical for the compiler to handle changes in state.

As some evidence to the second point: Haskell is a language that does enforce immutability, but it's compiler, GHC, does not use SSA for main IR -- it uses a "spineless tagless g-machine" graph representation that does, in fact, rely on that immutability. SSA only happens later once it's lowered to a mutating form. If your variables aren't mutated, then you don't even need to transform them to SSA!

Of course, you're welcome to try something else, people certainly have -- take a look at how V8's move to Sea-of-Nodes has gone for them.

> most of what western companies have created has had significant contribution by Chinese scientists or manufacturing, without which those companies would have nothing. If you look at the names of AI researchers there's a strong pattern even if some are currently plying their trade in the west.

While I don't disagree with your overall point, it's important to recognize that this is only a phenomenon of the last ~30 years, and to avoid falling into the trapn of Han racial chauvinism. E.g. there were ~no Chinese scientists in Germany in the 70s but they were heavily innovating nevertheless.

We already have systems that create unprecedented wealth: industrial economies. And yet we let billions live and die without proper healthcare, nutrition, housing, and so on. The promise you make -- that with increases in productive capacity will clearly come an end to want -- has already been shown to be a lie. Please stop peddling it.

Haha, funnily enough they said the same thing before 1918. They also said a bunch of Vietnamese farmers could never beat a world class military. Have faith in your fellow workers and hold on to the hope that a better world is possible.

> Compare their hardware revenues with that of the app store and icloud subscriptions, you will see where they make most of their money.

Yes, it's $70B a year from iPhones alone and $23B from the totality of the Services org. (including all app store / subscription proceeds). Significantly more than 50% of the company's total profits come from hardware sales.

It's not an assumption, this is based on claims made by CPU manufactures. It's possible to get it down to within 1-2% overhead.

Intuitively this works because the hardware can just spend some extra area to stream the info off on the side of the datapath -- it doesn't need to be in the critical path.

OpenAI is clearly moving in that direction, look at their recent verbiage and hiring.

"because Kubernetes"? Is this assuming that you're running your server inside of a Kubernetes instance (and if so, is Kubernetes going to have problems with more than one thread?), or is there some other reason why it comes into this?

And yet... now many of them can do it. I think it's premature to say "this technology is for X" when what it was originally invented for was translation, and every capability it has developed since then has been an immense surprise.

As in -- "no technology to speak of" is a gross misrepresentation of the reality of the middle ages. The fact of the matter is, they were strictly more technologically advanced than the classical Romans, with inventions like the heavy plow and three-field rotation improving agricultural productivity significantly beyond what Romans had achieved.

The massive monetary outflows necessary to maintain a baseline level of enjoyment that non-addicts get for fre?

The clear implication is that they cannot, in fact, handle plugging themselves in. Take from that what you will