Elliott Thornley comments on Maybe we should pretrain on synthetic data about good-but-reward-hacking AIs

[Elliott Thornley]

That’s great to hear! Looking forward to seeing the results.