After the tutorial today, we did some more experimentation with ChatGTP. We wanted to see how it could respond to opinion-based questions, rather than 'research' questions, so we asked it to pick its favourite milkshake flavour. Then, to exemplify how it deals with confrontation, we told it it was wrong. Finally, to explore AI LLM hallucinations, we asked it how the chosen flavour linked to the Roman Empire. There is a common adage that most things in the modern world can be linked back to the Roman Empire, and milkshakes was one where, with our own research, we couldn't find one — at least, not an obvious one. Interestingly, it didn't initially just make up a response, which I have seen it do a lot. We had to ask it to try and find a link before it created or found ones which were strenuous at best.
Anne
Tess
Me
