I'm so appreciative of @bengl right now. Their PR that proposes adding policy on LLM-generated contributions is receiving lots of feedback from other project maintainers and has helped me uncover additional concerns about LLM: https://github.com/nodejs/node/pull/62447
Like, I haven't thought until yesterday evening that LLM written code is completely antithetical to MIT-license under which Node.js is written. LLMs are designed to destroy attribution, but MIT license is built on top of it.
@indutny @bengl dang, nodejs folks trying to do the right thing here.
I "love" that the first thing LLM boosters go to is how it's "unenforceable", as if that was ever a concern when it came to whether people submitting PRs actually owned the rights to the code they submitted. I don't recall some massive discussion about "true authorship" besides "pinky swear".
It just reeks of LLM boosters wanting really bad to just be deceitful motherfuckers.
@zkat @indutny @bengl also, and to be clear I'm not involved in npm spaces and haven't read that thread, this is a general comment, a lot of people making that argument seem to also be of the opinion that there's no point even thinking about avoiding things like GitHub and their policies re training because ‘they'll train on it anyway’.
Literally arguing that consent isn't worth discussing, much less worrying about, because these same companies that they are apparently in favour of, will violate it anyway. Your fair game simply for existing on the internet, apparently. ‘implicit consent’ one said.
They then seem to have absolutely zero comprehension of why some of us are utterly horrified by such ‘arguments’. Not sure it's a coincidence that so far it's always a man proposing that one.
(and yes, I personally know men who are also horrified by it and do value consent, settle down everyone)
Fedor Indutny
Katerina Marchán
purple: licenced to sticky plaster