I'm not an expert on question-answering natural language processing algorithms, but it is a very dynamic area of research, and you can easily Google it, to find many blog entries and papers on recent advancements in this area.
Since I don't have access to any such algorithm, let me use example of text-generating algorithm based on the, state of art, GPT-2 model (it can be also adapted for question-answering as far as I know). You can check it online. Below you can see prompts (bolded) and the generated text that I've chosen from several attempts.
Today is Sunday. I'll finish my project in two days. I will finish my Project at the end of the week.
Inbetween, I'll be working on my Monday Job. [...]
or another one
Today is Sunday. I'll finish my project in two days. I will finish my Project at 5 a.m. on Monday, October 21st.
How do you deal with changes in time and personality?
As you can see, both generated texts show that the model does not understand the prompt and the idea of time. While such algorithms are pretty good at what they are doing, they are still pretty dumb and what they learn, is they memorize and learn to find patterns in the text, to complete the sentence (or answer, that's the same task, just you need different data to train the model).
