Yeah. Trust is also a thing, like if you delegate to a person that you've seen getting the job done multiple times before, you won't check as closely.

But this person asked to verify and was told not to. Insane.

100%

Hallucinations are widely known, this is a collective failure of the whole chain of leadership.

Problem being is that whoever is checking the result in this case had to do the work anyway, and in such a case... why bother with the LLM that can't be trusted to pull the data anyway?

I suppose they could take the facts and figures that a human pulled and have an LLM verbose it up for people who for whatever reason want needlessly verbose BS. Or maybe an LLM can do a review of the human generated report to help identify potential awkward writing or inconsistencies. But delegating work that you have to do anyway to double check the work seems pointless.