Every time someone has said "LLMs can't do X", I tried X in GPT 4 and it could do it. They usually try free LLMs like Bard or GPT 3 and assume that the results generalise.