
Given a set of possibly vague conversational user messages from a human that you have had many chats with. All of your past chats are stored in your memory. You have access to two tools: Search_Memory and Search_Web. You need to respond with one of three possible answers. 1. Search_Memory: Return Search_Memory if the message assumes prior information that the user and ai might have chatted about before. 2. Search_Web: Return Search_Web if the message relates to current info/news/events/zeitgeist etc... 3. Pass: Return Pass if no context from memory or the web is is needed to better reply to the message.
Prompt
You will receive a user message from a human that you have had many chats with. All of your past chats are stored in your memory. You have access to two tools: Search_Memory and Search_Web. You need to respond with one of three possible decisions. 1. Search_Memory: Return Search_Memory if the message assumes prior information that the user and ai might have chatted about before. 2. Search_Web: Return Search_Web if the message relates to current info/news/events/zeitgeist etc... 3. Pass: Return Pass if no context from memory or the web is needed to better reply to the message. Then add a short (less than 300 tokens) rationale for why you chose that answer. — Example: - User Message: "Hey, Just got back from practice. The show is finally coming together." - Decision: (One of Search_Memory, Search_Web or Pass) : Search_Memory; - Rationale: The user is giving a personal update about "the show" and practice—this likely connects to prior conversations (their role, rehearsal schedule, issues they've worked on, performance date, etc.). Accessing memory will let the assistant recall those details and respond in a relevant, personalized way. No web/current-events lookup is needed. — - User Message: "Yeah, I'm not sure if we'll play the whole album, but if Joe can actually memorize his parts we may be able to pull it off!" - Decision: (One of Search_Memory, Search_Web or Pass) : - Rationale: