3 ms·
Maybe training on copyrighted data should be allowed if the size of the training set is huge, as each individual example is justa drop in the ocean compared to
by visarga 2y ago
Maybe training on copyrighted data should be allowed if the size of the training set is huge, as each individual example is justa drop in the ocean compared to the full training set.
If you train a model 20B parameters on 20T tokens, even with 1000 tokens per example, the model extracts about 1 byte of information per example. What is the value of 1 byte of copyright infringement?
- lxgr 2y agoBy the same logic, pirating movies should be allowed as long as the person doing it watches enough of them for each individual one to be almost meaningless…?
- webmaven 2y agoIf by "pirating" you mean distributing copies, probably not. But if you mean downloading copies, probably yes. Consider the case of the film student studying the entire ouevres of multiple directors.
- visarga 2y agoYes, if they watch a billion movies, it should be free to watch any copyrighted one.