Jack Hopkins

Over­think­ing: Am­plify­ing rea­son­ing weights makes mod­els re­veal their secrets

Self-At­tri­bu­tion Bias: When AI Mon­i­tors Go Easy on Themselves