A short/summary/long-term memory structure and fact verification system reduce setting breakdown in long chats.
A prompt chain is designed so the model asks questions rather than guessing when it is unsure, reducing hallucinations.
Last updated 2 months ago
