LLMs fail in ways humans never would | Dark Hacker News