TurnTrout comments on TurnTrout’s shortform feed

[TurnTrout]

Transparency Q: how hard would it be to ensure a neural network doesn’t learn any explicit NANDs?