True, virtually all companies harvesting LLM training datasets don't bother honouring even the most permissive licenses, like MIT or BSD - Microsoft leading the pack with Github and Copilot.

You're right to point out that the tide is shifting again. Perhaps at the end of this bubble, society and/or the behemoth companies will recognise the value and help build a more sustainable future for artists and creators. I'm cautiously optimistic.