Notes & essays.

A Safety Check for Deceptively Aligned Mesa-Optimisers

An incentive based safety check for superintelligent AI that may exhibit inner misalignment

• ai
• ai-safety
• alignment
• deceptive-alignment