> If AI can't use ideas from copyrighted works, then nobody can. Because anyone could be secretly using AI we have to apply the same strict standards of attribution to all human works.
Yes, we should be applying strict standards of attribution to all works, human or otherwise.
IMO one of the biggest defects of language models currently is lack of attribution; if sources are not properly cited, there is no way to verify the veracity of what is written. Unfortunately, apart from the technical difficulty of implementing attribution, the exact combination of sources is part of a model's "secret sauce", so there is an economic incentive to not cite.
Yes, we should be applying strict standards of attribution to all works, human or otherwise.
IMO one of the biggest defects of language models currently is lack of attribution; if sources are not properly cited, there is no way to verify the veracity of what is written. Unfortunately, apart from the technical difficulty of implementing attribution, the exact combination of sources is part of a model's "secret sauce", so there is an economic incentive to not cite.