> Mathematics is about discovering and understanding the logical implications of assumed axioms under various inference rules.

That's like saying that programming is about producing valid programs in various programming languages.

It is though. I am not being edgy. Programming is about making valid programs, yes. I literally have no idea what else you might be gesturing towards.

Distillation of our blessed models is no fair!

> Also in a TUI you never ever have that issue that some text isn't copy/pastable like some dialogue boxes

May I present OpenCode, which somehow managed to break copy from terminal: https://github.com/anomalyco/opencode/issues/4283

Self-hosting won't protect you from getting locked out of a closed weights model, because you can't self-host it even if you have the hardware.

If they "should" cost 4k in the sense of marginal cost, then you will be spending more running the same at home, because your home hardware will always be less efficient.

There is a big difference in the cost of a 5-nines up time system in a heavily space constrained environment compared to a home hobby white box used for some coding. The GPUs alone cost 10x for the data center versions compared to the gaming versions even with similar specs.

The cost of online services is also largely a result of the cost of training (though hard to say exactly what that number is). Assuming you are using open weight models at home, you aren't paying for the training - someone else is.

> The cost of online services is also largely a result of the cost of training

OpenRouter prices are somewhat simmilar to Antrhopic/OpenAI API prices. So I conclude that the hardware plus operating margin alone can genuinely produce prices way above what you'd pay if you had a subscription. Of course the primary unkown factor is average token use per subscription. Without that it's all wild speculation.

As of now, there are only 2 natural language clusters in the partition: Claudish and Spanish/French. From a first glance, the other clusters seem centered on technologies (lots of command line flags and abbreviations among the "most representative" words). If the goal is to investigate English language usage trends, it might be better to cluster based only on English words.

Their suggestion to tax tokens extends the problem of taxing wages to a new type of production. The real solution is, of course, taxing the capital, but you will never hear that from them.

The bots can also write Rust instead of C++, doing away with the arcane nonsense accumulated by that legacy language. SCNR.

Bot-assisted Rust could be amazing; there's some ports already happening which wouldn't have otherwise. Maybe Rewrite-it-in-Rust can actually be a real thing and not just a meme. But it does put a big burden on implementors to understand what they're generating, and now it's in an unfamiliar language to boot.

Bureaucracy gives you leverage against slop. Review seriously, but limit the time that you spend. This will stall the slop. When the culprit complains, tell your boss "I spend X hours per week on reviews. If you need more throughput, the PRs quality needs to improve."

Writing and reading design documentation can be slower than pair programming. On the other hand, info about code design also belongs into inline documentation or commit messages (in this order of preference), so the effort might not be wasted.

I don't think so. There can be a lot of shared context within the team which can make prose shorter than writing code. And written words last longer than verbal exchange.