There's another very simple test you can put ChatGPT through: Ask it to multiply some small numbers which it will most likely get right. Then ask it to multiply some large numbers and it will most likely be wrong by an order of magnitude or more. My layman's explanation is that because the space of likely answers is much smaller for small numbers than for larger ones the probability of finding the same multiplication in its corpus is high and the result likely correct. The larger the numbers the less likely it is to find the same multiplication or find an inference (for example both numbers begin with a 2, the result is likely starting with a 4) that is correct for all digits.