Comment by ben_w

> Can vibe coder gurus create operating system from scratch that competes with Linux and make it generate code that basically isn’t Linux since LLM are trained on said the source code …

No.

Vibe-coding, in the original sense where you don't bother with code reviews, the code quality and speed are both insufficient for that.

I experimented with them just before Christmas. I do not think my experiments were fully representative of the entire range of tasks needed for replacing Linux: Having them create some web apps, python scripts, a video game, a toy programming language, all beat my expectations given the METR study. While one flaw with the METR study is the small number of data points at current 50% successful task length, I take my success as evidence I've been throwing easy tasks at the LLM, not that the LLM is as good as it looks like to me.

However, assume for the moment that they were representative tasks:

For quality, what I saw just before Christmas was the equivalent of someone with a few years' experience under their belt, the kind of person who is just about to stop being a junior and get a pay rise. For speed, $20 of Claude Code will get you around 10 sprints' equivalent to that level of human's output.

"Junior about to get a pay rise" isn't high enough quality to let loose unchecked on a project that could compete with Linux, and even if it was, 10 sprints/month is too slow. Even if you increase the spend on LLMs to match the cost of a typical US junior developer, you're getting an army of 1500 full-time (40h/week) juniors, and Linux is, what, some 50-100 million developer-hours, so it would still take something like 16-32 years of calendar time (or, equivalently, order-of 1.2-2.5 million dollars) even if you could perfectly manage all those agents.

If you just vibe code, you get some millions of dollars worth of junior grade technical debt. There's cases where this is fine, an entire operating system isn't one of them.

> Also all this on $20 plan. Free and self host solution will be best

IMO unlikely, but not impossible.

A box with 10x the resources of your personal computer may be c. 10x the price, give or take.

While electricity is negligible (which today, hah!): If any given person is using that server only during a normal 40 hour work week, that's 25% utilisation rate, therefore if it can be rented out to people in other timezones or where the weekend is different, the effective cost for that 10x server is only 2.5x.

When electricity price is a major part of the cost, and electricity prices vary a lot from one place to another, then it can be worth remote-hosting even when you're the only user.

That said, energy efficiency of compute is still improving, albeit not as rapidly as Moore's Law used to, and if this trend continues then it's plausible that we get performance equivalent to current SOTA hosted models running on high-end smartphones by 2032. Assuming WW3 doesn't break out between the US and China after the latter tries to take Taiwan and all their chip factories, or whatever