I broadly agree with you, but I don't see what's contradictory about the solution of model weights becoming public domain.
When it comes to piracy, the people who have viewed it as ethical on the grounds that "information wants to be free" generally also drew the line at profiting from it: copying an MP3 and giving it to your friend or even a complete stranger is ethical, charging a fee for that (above and beyond what it costs you to make a copy) is not. From that perspective, what OpenAI is doing is evil not because they are infringing on everyone's copyright, but that they are profiting from it.
When it comes to piracy, the people who have viewed it as ethical on the grounds that "information wants to be free" generally also drew the line at profiting from it: copying an MP3 and giving it to your friend or even a complete stranger is ethical, charging a fee for that (above and beyond what it costs you to make a copy) is not. From that perspective, what OpenAI is doing is evil not because they are infringing on everyone's copyright, but that they are profiting from it.