Bug:
The model tends to treat quotes from another person as the user’s response.
I mean, when I talk to the model and quote something else from someone, there’s a high chance it will treat that as my dialogue, not a quote. Even when I include “”, it still has a high chance of treating it like my response.
Hello,
You may need to explicitly specify that you are quoting someone in your prompt so the model can correctly interpret it. Since LLMs are autoregressive models that means they predict the next token based on the previous ones, they might not treat text as someone else’s quote unless it is clearly indicated.