Comment by pm215
18 days ago
That's funny, but also interesting that it didn't "sign" it. I would naively have expected that being handed a clear instruction like "reply with the following information" would strongly bias the LLM to reply as requested. I wonder if they've special cased that kind of thing in the prompt; or perhaps my intuition is just wrong here?
A comment on one of the threads, when a random person tried to have copilot change something, said that copilot will not respond to anyone without write access to the repo. I would assume that bot doesn't have write access, so copilot just ignores them.
AI can't, as I understand it, have copyright over anything they do.
Nor can it be an entity to sign anything.
I assume the "not-copyrightable" issue, doesn't in anyway interfere with the rights trying to be protected by the CLA, but IANAL ..
I assume they've explicitly told it not to sign things (perhaps, because they don't want a sniff of their bot agreeing to things on behalf of MSFT).
Are LLM contributions effectively under public domain?
IANAL. It's my understanding that this hasn't been determined yet. It could be under public domain, under the rights of everyone whose creations were used to train the AI or anywhere in-between.
We do know that LLMs will happily reproduce something from their training set and that is a clear copyright violation. So it can't be that everything they produce is public domain.
This is my understanding, at least in US law.
I can't remember the specific case now, but it has been ruled in the past, that you need human-novelty, and there was a case recently that confirmed this that involved LLMs.