Punishing AI for lying and cheating might not be such a good idea after all
www.livescience.com
Scientists at OpenAI have attempted to stop a frontier AI model from cheating and lying by punishing it. But this just taught it to scheme more privately.