x
The Self-Diagnostic Limitation: Why AI Systems Can't Reliably Assess Their Own Alignment — LessWrong