2 ms·
So what? If training on copyrighted data without authors consent is ok then distilling is ok as well.
by yreg 3mo ago
So what?
If training on copyrighted data without authors consent is ok then distilling is ok as well.
- nonethewiser 3mo agoMaybe both are ok. Maybe neither. Maybe one. The point is this does not follow >If training on copyrighted data without authors consent is ok then distilling is ok as well. Because building a model from distillation and building a model from raw data are not the same. You have to evaluate them independently. And legally its different as well. IP vs ToS (civil).
- yreg 3mo agoNo, it follows because the AI labs can at best argue that some copyright over the model output was broken. They can hardly successfully sue a rival for breaking ToS.