As I’ve interpreted it, they mean that LLMs and similar tech isn’t particularly helpful to theorem provers at the moment.