Surface level solutions (which work today) would involve being more strict about commercialisation rights, e.g. in Copyright. If you train on web-scraped data, that's OK for research, personal use and non-profit only — but you *cannot* commercialize.

→ View original post on X — @alexjc