I'm unsure that using complicated algebra to regurgitate parts of people's works is different from just copying it. Perhaps you could say a human brain learning how to code is just regurgitating the code it's had as an input before, but intuition says directly copying is somehow different.
I add copyleft licenses to code to ensure the code cannot be legally copied into proprietary software, for moral beliefs. If the output of a LLM was free software and copyleft (as would be the input) then perhaps that would be fine. Github probably has some complicated legalese that says by uploading it you permit them to use it for LLMs - I'd want that to be legally voided.
The purpose of copyleft is to ensure the freedoms and restrictions made by a license are enforced by copies/forks/derivatives.
Best to talk to people interested in contributing to see what they would think?