What ppl often refer to as 'AI' these days are LLMs (large language models), which are essentially text-generators (like an advanced version of the predictive typing feature on your mobile phone)...

LLMs can't do math and they can't count, they just generate text that sounds/looks right based on training data - a classic example is chatgpt stating there are only two 'r's in 'strawberry'

An old date-time library would be vastly more accurate than 'AI' in computing date intervals