It's interesting that despite how common polyamory is in SV, even "eccentric" tech billionaires have to be presenting as monogamous.

Polyamory still comes off as a low status behavior.

Every time see this story I think "oh, this is the story about the packet TTLs being set stupidly low or something but you wouldn't be able to narrow that exactly to 500 miles" and have to click and learn again the the first time it's about the connection timeout being set stupidly low.

This is almost the origin story for the EDM producer deadmau5's name.

It's only fair you ask an LLM to review it for you.

Yann joins Ilya, Karpathy, Sutton + Carmack when he says LLMs are a dead end, though.

Karpathy is probably the most careful not to write off LLMs entirely but he seems pretty skeptical.

Once you accept Curry-Howard, untyped FP languages are hard to take seriously as a foundation for reliability. Curry-Howard changes the entire game. FP and strong types were clearly meant for each other.

Untyped FP languages can be productive, flexible, even elegant (I guess) but they are structurally incapable of expressing large classes of correctness claims that typed FP makes routine.

That doesn’t make them useless, just, you know. Inferior.

To paraphrase Bjarne Stroustrup, there are two kinds of programming languages. There are abominations and then there are the ones nobody uses.

I'm impressed Torvalds managed to not know what he was referring to (the Twitter firings).

The missing context whenever this comes up is the fact that it was a surprise one off.

If developers have no idea they're going to be graded by lines of code at some random future date that's a much different situation than saying you're going to give bonuses away every month based on how many lines of code were written.

Everyone knows the second is bad, it'll be gamed massively. The first one could be useful though.

And yes doing it as a one off is still problematic and you can think of all kinds of exceptions, but if you think the organization is full of dead weight in general and overhired massively, a crude stack ranking by lines of code is a pretty good metric for figuring out which (e.g.) 50% is the bottom.

> a crude stack ranking by lines of code is a pretty good metric for figuring out which (e.g.) 50% is the bottom.

I can write you an efficient algorithm in 2 lines or an inefficient one in 50. The metric is about as useful as a doctor checking how often someone picked up a bottle to figure out how much they drink.

> I'm impressed Torvalds managed to not know what he was referring to (the Twitter firings).

I mean, naughty old Mr Car didn't _invent_ this nonsense; IBM was fairly notorious for it in the 80s, say. He's probably the most prominent recent example.

> The first one could be useful though.

How?

> a crude stack ranking by lines of code is a pretty good metric for figuring out which (e.g.) 50% is the bottom.

No. It's really not. For a start, you probably lose basically everyone very senior by that mechanism. But also you lose the troubleshooters.

Torvalds talking out of his ass? impossible

Last April I asked Claude Sonnet 3.7 to solve AoC 2024 day 3 in x86-64 assembler and it one-shotted solutions for part 1 and 2(!)

It's true this was 4 months after AoC 2024 was out, so it may have been trained on the answer, but I think that's way too soon.

Day 3 in 2024 isn't a Math Olympiad tier problem or anything but it seems novel enough, and my prior experience with LLMs were that they were absolutely atrocious at assembler.

https://adventofcode.com/2024/day/3

Last year, I saw LLMs do well on the first week and accuracy drop off after that.

But as others have said, it’s a night and day difference now, particularly with code execution.

Yes. Anyone who thinks you can ship a datacenter to space and save has never managed a datacenter.

Or worked in space.