martinkunev comments on Testing for Scheming with Model Deletion

[martinkunev]

I wrote something on a related idea a while back:

Disincentivizing deception in mesa optimizers with Model Tampering