It's ironic that the one thing LLMs can't do reliably in this space is "write copy for humans" (I don't trust them for that either).
Kind of funny to do something impressive and then ignore the details on the presentation, but perhaps that's not uncommon for security researchers?