> Yea it feels misleading. I think finding attribution is genuinely a hard problem and a solution in this space would be valuable.

Wouldn't the attribution be basically all the dataset with various attribution probabilities? It's kind of like a reverse neural network.