@doug I also don't see how relaxed views on intellectual property are "anti-socialist". I am deeply opposed to the US system of IP law, especially with recent attacks on fair use. I don't think that artists should be able to retain complete and permanent ownership over their works once they are available to the public. I believe that information should be free, and that expression, even controversial expression, should not be restricted by anyone's financial interests.
@doug FWIW, The USSR had copyright laws that lasted 15 years. If we had similar laws, training a model on only public domain materials would be extremely easy.
But in the USA, thanks to corporate lobbying, a copyrights lasts 70 years, and a trademark can be renewed indefinitely. Narrowing datasets down to the public domain is impossibly complicated.
What choice do AI engineers have but to ask for forgiveness rather than permission in these matters?