Gary Marcus Says LLMs Fail at Following Rules

Gary Marcus, identified in the packet as a cognitive scientist, author and critic of deep learning, posted a reply on X. He wrote: "no and that’s the fucking problem. LLM-based systems just aren’t good at following rules." The statement was made in response to a tweet discussing the addition of ethical instructions to prompts. Marcus’s comment rejects the notion that such instructions can reliably prevent safety incidents. The packet presents the remark as a direct quote from his account and supplies no independent confirmation or further details.