Here's another one.
This is a classic logic puzzle - usually about ducks.
There are two pineapples in front of a pineapple, two pineapples behind a pineapple and a pineapple in the middle. How many pineapples are there?
When you use ducks, Gemini can do it, when you use pineapples it cannot and thinks there are 5 instead of 3.
ChatGPT 3.5 and 4 can do it.
The even funnier thing is if you then say to gemini, hey - would the answer be the same if it was ducks? it says NO then there would be 3 ducks and explains why.
Then if you say, but wouldn't the same logic apply to pineapples? It says 'oh yeah!'.
Anyone saying I am just playing with a silly thing, I say no - this is testing reasoning ability and understanding which is the number one thing I want.
ChatGPT appears to have a more generalised ability to reason whereas Gemini is only reciting its training data.