> LLMs are literally making up stuff through statistical linguistic inference on a grand scale from hearsay.

Hearsay being some personal "truth" it may be useful to know what the statistical average of that "truth" is. If we do it right perhaps we can get the various personal errors to cancel out.