In my project I developed a group chat where my GPT participates actively together with several users.
API call example:
user: user1 said: …
user: user2 said: …
assistant: assistant replied: …
user: user2 said: …
The problem is that other participants may talk to each other.
In my current implementation, the bot answers only when someone calls its name but it would be interesting to let the AI decide when to answer and when to not answer.
A possible solution would a clever porompt that makes it respond with a special phrase in case no answer is required from the bot and then my middle layer blocks this response from showing in the chat.
But this solution does not sound robust.
I wonder if others already solved a similar problem or any other ideas.
Thank you