momom2 comments on Alignment can improve generalisation through more robustly doing what a human wants—CoinRun example

[momom2]

Nice! Is this on ai-plans already?