Reasoning Models Sometimes Output Illegible Chains of Thought | Read Paper on Bytez