Charlie Steiner comments on Alignment can improve generalisation through more robustly doing what a human wants—CoinRun example

[Charlie Steiner]

What is the algorithm?

[rokosbasilisk]

not much information is given regarding that so far, i was curious about that too