Why is it so difficult to expect the LLM to process and pass date alias such as today, tomorrow, next week, last year?
It is easy to expect. It is difficult to receive.
Large Language Model AI is a text predictor. The mini-sized model isn’t a very good predictor.
AI has to know today’s date to come up with tomorrow’s date. Are you providing that by API?
AI has to, ideally, observe a calendar to understand how many days takes it into next week, what the days of the week are even into past or future months, and make the minimum required computation-like prediction, as it is not a date calculator that can know October 27, 2024 is a Sunday.
(date range above is in error - perhaps “next week” is most likely three days out…)
Or might even need to know the time, or more information in general about the consultants not working on Sunday or being available late.
So with development work, providing context, providing calculators, providing background, you can produce an application. A link to a scheduling web site would be a better AI product here for the examples I show.
1 Like