Comment by ghostpepper

Any time someone shows me something an agent did, the first question I ask is whether there's plausibly something comparable already in its training data.