I think the HOW and WHY here are really important:

By incorporating copyleft data into their models, the LLMs do share the work - but not alike. Instead, the AI strips the work of its provenance and transforms it to be copyright free.

That is because the US copyright office has rightly advised that AI generated works are not copyrighted.

So crucially, here, we have them gobbling up methods and ideas and passing them off as their own, stripping the code of its copyright protection.

There's a whole other argument to be made about whether the cod actually works, but that feels too close to being apologist for me.