I have found LLM AI systems will stand firm on unverified assumptions while sounding confident and reasonable.

It appears the programme can only predict the most likely sequence of words rather than dynamically verifying the truth.

I conclude they are engines for generating plausible text, not infallible sources of ground truth and must be treated as such. One would be wise to consider their output as a draft to be carefully cross-checked before acting.