Language models can learn to tell when their activations are being read

1 points | by torsi0n 6 hours ago