not producing precisely what the content creator wanted, so it favors people who don’t really have an opinion on the specifics

That's an interesting point that I haven't run into elsewhere!

All C code can be compiled with the C++ compiler, though there's some nuance (if you don't mark the C code as C, it behaves slightly differently which could mean it doesn't work).

C++ adds a bunch of language features (object-oriented programming and templates are two massive ones). This can make it easier to describe a concept in C++ than in C. But it also enables a lot of unneeded complexity (if C++ wasn't the best fit, or if the programmers aren't excellent) and the compiler is slower because of all the extra stuff. And the resulting binaries are much bigger. And it's harder to port the C++ compiler to other platforms. There's still plenty of situations where it makes more sense to use C over C++.

2d->3d -- bigger number

c -> c++ -- two extra plusses (and plus already means extra!)

games -> epic games -- it's got "epic" in the name

You don't have to be a genus to see these are all better!

I really appreciated your explanation; it was clear and I think you're correct. It's disappointing that Linus is either unaware of it or doesn't care about deterministic results.

Is that what Linus is talking about in the OP? I was sure this was about code-generation. If he's talking about images and recording the metadata then I agree with you.

You make a good point about capability. However, I'm trying to say that in practical capability, I do know exactly what compiler flags were enabled (it's whatever the Makefile passed, which is whatever configure determined, the whole thing is checked in as source code), but the exact model+seed isn't being recorded for this Linux changes.

To make the compiler analogy, we should be saving model, seed, prompt, etc and then the LLM output becomes the equivalent of the compiled code. I think that's an extremely defensible argument, while "bet you can't write assembly!" is not.

I'm kind of gobsmacked by this. I've heard younger developers compare LLM with compiler and chalked it up to ignorance, but I don't get what this is.

Can I write code in assembly? Yes. Of course. It's true that I never do it, but if there was a rule that you needed to write some assembly now and then to certify as a kernel developer I don't see how it would be any kind of impediment.

Comparing more directly: if you hand me a compiled binary and the source code and ask "why is the bit at address 0xXXXXXXXXX a 0?" I can, with 100% certainty, answer that. Probably in less than a day of effort. If I ask "why did the LLM output "for" as the XXXXth word of its utterance, can Linus say why?

And "can you write a good commit message without an editor?" Uhhh? At least when I was in school, this was something people did for fun/messing around. I still do cat > somefile when I need to put some quick text together. In fact, Claude will sometimes write a Git commit by cat-ing into a file--which it's doing because there's so many online examples of this. Fucking weird things for Linus to pick. Makes me think his brain's rotted.

Endless Sky is excellent. It captures all the enjoyment of the EV games, and its free for everyone.

Your controller looks great. Just the kind of thing an independent ES pilot would be flying with!

Pretty neat that the advertisement has "or buy our materials and make them yourself!" in two different places.

Yup. I've been in academia and in private business. Academia pays terribly and expects a fuck-ton more work. I'm sure there's places where you can just sleaze by without doing much, but I never saw them. (Which isn't to say that all academic work is good work. I saw people who got far doing poor research, but they were also working very hard at it.)

Those sound like bad-ass professors!

I think "support the working class" doesn't have to mean "do whatever they say". We're all generally propagandized into oblivion. Like, the US working class has been concerned about immigrants and trans women in sports but it's not pro-worker to go after those things.

I have some similar feelings to AI here. The advantages of self-checkout are only present because we've made existing systems worse to the point where something not-great is now a better option.

In the 1980s, a grocery lane would have a check-out clerk and a bagger. Sometimes two baggers. You would put your items on the belt and they'd be scanned and bagged (well, so that eggs weren't crushed, freezer items were together, etc) about as fast as you took them out of the cart.

By lowering wages and reducing staff, this got worse. Now the check-out clerk is poorly-trained and probably quitting/being laid off soon and the bagger has been laid off. Mysteriously, the human-checkout line is now kind of slow and frustrating!

Having individual customers do the check-out themselves is even worse (so frequently, someone gets the machine stuck yelling about putting items in the bagging area, or just stares vacantly at the screen), but by parallelizing the process it comes out to usually slightly faster.

The companies won’t pay the 10% out of their own pocket, as they most likely couldn’t. At least here in Germany, supermarket margins are pretty thin.

This doesn't pass a quick reasoning check:

1. Self-checkout machines cost the store capital investment (they introduce some new technology, there is a larger number of devices)
2. Self-checkout machines cost the store operating budget (more devices means more parts to be serviced)
3. Stores have added significant numbers of self-checkout machines

So the stores must be getting better margins at self-checkout than at human-checkout, or they wouldn't introduce it.

The problem is that they can't just talk about climate problems they need to take action. And since the Democratic party isn't going to support mass-executions of industry executives and billionaires, they'd have to talk about incremental changes. I'm not sure how they sell something like "we forecast that these new regulations will result in a 1 degree average temperature reduction over the next 4 years" (which would be huge). And that's imagining that the party as a whole is able to agree on something hard enough to shove it past the Republicans.

I think what you're saying here is that by "it's not possible" you meant "it's not possible to hold companies liable for their chat-bots' words and still have those companies producing chat-bots." I agree. I think that would be the correct outcome.

There would be some very narrow spaces where LLMs would be usable, but mostly they would cease to pollute everything. Unrelated to AI, I'd like to see lots of regulation that blocks externalizing costs. For example, maybe it's also "impossible" for Walmart to operate in the US if they don't have all their employees getting state assistance. That would also be good. Maybe it's "impossible" to have a beef industry without massive pollution and deforestation. Also good. Et cetera.

How would that be technically impossible? Or difficult? Companies are currently liable for the actions of the human agents, and I don't think the implementation needs to be different.

Maybe you mean socially impossible? Like, the major providers have a lot of money behind them and so why would they allow themselves to be liable for anything?

Companies should be liable for what their LLM says.

Yeah it really got the Batman atmosphere well. I wonder if it's because Batman was such a cultural icon that even if you "didn't read comics" you till knew who Batman was and what he was about?

Who wrote these regex? They are terrible:

• We've got a capture-group for the numbers (the () means we're saving that part for later). Are we planning to do something with the mlg numbers? Why don't we care about the mmmberry numbers?
• The capture is one or more numbers, but then it's an optional group. Write ([0-9]*) instead
• We match mlg and mlg1 but not mlga because we anchor to the end ($). But not on mmmberry--that one will match "mmmberry5" (but we don't track the number!) but also "mmmberry_is_the_worst" so the enemy of my enemy is also my enemy because they share a prefix

Maybe mlg has sock-puppet accounts mlg1, mlg2, but never MLG?