ChatGPT is, in it’s core, a sequence predictor/generator. Give it some context and it will produce most probable results.
Having the context of man, boat and river is very likely to match this “how does the man get the goat on the other side of the river” riddle, which is probably very predominantly present in the data set ChatGPT was trained on.
That’s why ChatGPT can be helpful for lots of different purposes, but shouldn’t be taken literally. Errors are very likely to happen. It can give important hints, but always check the results thouroughly.
Thanks for pointing it out. There is clearly room for a lot of error.