This was submitted by a bot :D I subscribe to Fabien's RSS and he must have changed something in the server because I got this post on my RSS reader (again, as it's an old post), and here it is submitted to HN (again)
The promise of AI gains with fewer people is what's triggering this - before, you needed ~1 coder to do ~1 unit of work. Suits now think you only need 1 coder to do X units of work because LLMs.
They're laying off the people who can't produce a minimum of 2x with AI, and keeping the maximalists with no life outside of work barely keeping up with the 100k LOC a week they're shipping to prod.
Suits have an idea of what the New Model Coder should be, and it's not people who don't burn through 100,000,000 tokens a week.
Thanks for the blog post, interesting read. Question: why run `hugo serve` instead of using it to build the files and then use nginx or some other webserver?
We will at one point know more about the Roman Empire than the World Wide Web 1994-1998, and it doesn't seem right. Huge chunks of cultural output that would be gold for future historians just... lost.
> and their English mistakes are pretty uniform due to the specific differences between the languages, esp regarding prepositions and tense
This is not the case. Different people make different mistakes, as any exposure to a high-school level English class will allow you to discover. There's a case for clusters of similar errors, but then again any generalizing statement is true for a sufficiently high level of clustering.