“Current Language Models Struggle to Reason in Ciphered Language” by Fabien Roger | LessWrong (30+ Karma) | Podwise