

This is why I focus on distribution rather than training. If you commercialize a model trained on things you don’t own/license, and it generates anything remotely infringing, you should be fully on the hook for every single incident.
But if a model is trained and distributed freely as FOSS, then it’s up to anyone running it to ensure the output is not infringing. This protects fair use while also ensuring that big companies tread more carefully when redistributing models that can violate fair use by competing with those whose work was trained on without permission and are subsequently being emulated without permission.







Why do you care so much about defending unimaginably wealthy corporations stealing the labor of regular people?
See, now we have both misrepresented each others comments.