With all the hallucinating (or "bullshitting") going on, it's hard to imagine LLMs working well for query generation. But hey, we're _very_ early days for all of this.