The NLP detectTimeQuery piece that determines if users are making catch-up requests triggered when I asked the question 'How many alien sightings are reported per year?' This caused the LLM to refuse to answer, b/c it went through the timeWindow prompt rather than the semantic one. Could add more confidence score filtering here. I think it was only 0.4 in that case....
The NLP
detectTimeQuerypiece that determines if users are making catch-up requests triggered when I asked the question 'How many alien sightings are reported per year?' This caused the LLM to refuse to answer, b/c it went through the timeWindow prompt rather than the semantic one. Could add more confidence score filtering here. I think it was only 0.4 in that case....