Exactly this. In the right hands the outputs from LLMs are amazing. Not always right, but can be checked. In the wrong hands they are worse than useless because they cannot be checked.