Just a few links to complement Abram’s answer:On how seemingly myopic training schemes can nonetheless produce non-myopic behaviour:
Auto-induced distributional shift
Predict-o-matic
On approval-directed agents:
Tampering incentives of decoupled approval
Act-based agents
Just a few links to complement Abram’s answer:
On how seemingly myopic training schemes can nonetheless produce non-myopic behaviour:
Auto-induced distributional shift
Predict-o-matic
On approval-directed agents:
Tampering incentives of decoupled approval
Act-based agents