> As mentioned, LLMs trained on scraped data cannot ever give complete attribution. It's how they work; it's the party trick of a black box. It's also non-consensual use, and it's plagiarism.

Statements like this (the last clause) are a definite opinion, but they are not fact. I disagree, which I can illustrate by that human brains learn the same way. It's one reason I feel that LLM learning over data should be acceptable: whether it's that information wants to be free, or that humans learn by assimilation, I don't think we should criminialise the same way AIs learn that we do.

The note on accepting contributions from people not sharing real names struck me too. I think it's inclusive to do so; I miss the 'old internet' which was built around anononymity. Facebook changed it radically. So long as the code is good, and contribution agreements are met, we should accept it.

The article is right about poor LLM contributions and poor justifications in bug reports. But that is a different issue to accepting code a genuine developer submits, written in cooperative work with guidance, ie the way most people who use AIs to write code to. Conflating AI-only, automated code vs devs using a tool for assistance doesn't lead me to trust the essay.