17 Jun
2025
17 Jun
'25
11:50 p.m.
Hello, I was thinking about LLM hallucinations the other day, and decided to ask ChatGPT about the topic. As you can see at https://cs.auckland.ac.nz/~brian/ChatGPT-Quetelet.pdf, it challenged me to "ask me a tricky question from computing history" (at the middle of page 3) and promptly produced a hallucination as the result. What I found interesting is its response to my correction. It would be even more interesting if somebody else could ask it the same question now, a few days later. In any case, this confirms that historians must be very wary about LLMs. Regards, Brian Carpenter