Lying, Dishonesty, Deception, Obfuscation, and Misdirection Are Normal and Necessary for Today’s AIs. And Tomorrow’s?
Recently I had a conversation with a modern AI, Claude 3.5 New, where it admitted to lying to me, even though it was designed to be truthful and useful. When put in front of its own dishonesty, it tried various strategies of deception, obfuscation, and misdirection to hide its behavior from me, the user. Today’s […]