Anthropic explains how its Constitutional AI girds Claude against adversarial inputs

It is not hard — at all — to trick today’s chatbots into discussing taboo topics, regurgitating bigoted content and Leer mas

Via:: Engadget