I need some help from everyone. I just discovered that the chat log text for content queries in the knowledge base contains chain of thought content, which affects the accuracy of vectorized retrieval. I want to ask how to handle this. In theory, the think tag should have already collapsed the chain of thought. The chat window also shows it as already collapsed. I hadn’t noticed this issue before, so I’m not sure if it’s a new occurrence because today I suddenly felt that the content returned by the knowledge base was completely off, so I took a look. It was all good before.
