@iska@mk.starnix.network
Eh the copyright thing actually makes sense to me.

All methods I've heard of that allow credit-assignment to training data are computationally infeasible, but something should be done about that