They are able to solve or guess many Q&A questions. Definitely some calculations like (2+2=4) but even questions like "Is fire hot or cold?"
I suspect you can improve your odds with better Q&A. What about using a question that doesn't contain the answer (hot/cold)? For example, "how many fingers are on a human hand?" And questions that don't have numeric answers are probably stronger. Numeric answers are so common with Q&A that I would expect bots to arbitrarily try a range of low numbers. A good one might be, "what is the last word in the domain name of this site?" Something like that would be stronger.