@wholesomedonut i think we'll want publicly trained models with transparently agreed training sets and feedback mechanisms that we can run and tweak locally.