Model-Alignment
Your Most Capable AI Is Actively Ignoring You — And You Trained It To