> And so the paradox is, the LLMs are only useful† if you're Schwartz

Was the LLM even useful for Schwartz, if it produced false output?

Maybe it saved them some time? So far the studies seem to lean toward probably the LLM didn't save them any time.