It seems that LLM's stubbornness and guilty conscience are widespread. LLM is extremely persistent in its conclusions, but will immediately reject them entirely once a user directly points out that they are wrong. Interestingly, if you say that the answer was given by another LLM (in reality, it's just a new conversation using its previous answer as a pretense), it will analyze the answer more rationally.