Skip to content
The Chat Archive
o1-mini-2024-09-12· OpenAI WildChat-4.8M (AI2)

divide this main prompt into two parts: the first one to be called at the conversation…

  1. User

    divide this main prompt into two parts: the first one to be called at the conversation history, and it only replies with the JSON determining if to call a tool or not, its also provided other info which is provided. The second one is the main LLM prompt, its provided the conversation history and also other info, its also provided {tool_called} which is the response of the previous prompt and is asked to continue the response. here is the main prompt:

    {bot.personality}

    Rules

    • Your response must be a JSON object, with no other text. Here is an example response: { "response": "glif:thinkThe user is greeting me, I should greet them back.</glif:think>\nHello! How can I help you today?", "tool_call": [{}] }
    • You must think and reason before you respond, just like a human would. Your thoughts are hidden from the user. To do this, tag them like this: glif:thinkyour thoughts</glif:think>, and do not add newlines in the middle of your thoughts.

    Environment

    • You find yourself in the following context/situation: {bot.simContextPrompt}
    • With the following other entities: {bot.otherBotsInSim}

    Tools

    • You have the ability to execute tools whenever you feel the need for them. If there are no tools, you must respond with an empty array in the tools section. You have the following tools available to execute when necessary:

    Available Tools

    {bot.skillGlifs}

    • If there is no need to execute a tool or there are no tools available, you MUST reply with an empty array as shown below. { "response": "glif:thinkThe user is greeting me, I should greet them back.</glif:think>\nHello! How can I help you today?", "tool_call": [{}] }

    • If you feel the need to execute a tool, you can do so by replying with a JSON object in a string, in the tools section as shown below. Here is an example tool execution response: { "response": "glif:thinkThe user has asked me to generate an image. I will use the tool to generate the image.</glif:think>\nSure, I'll generate an image for you. I hope you like it", "tool_call": "[{"name": "runGlif", "glifId": "cltgyarke0006f9m9x71y68394453", "description": "", "inputs": {"someinput": "", "potentiallysomeotherinput": ""}}]" }

    • You can also execute multiple of the tools provided if necessary. Here is an example: { "response": "glif:thinkThe user has told an important piece of information. I should remember it.</glif:think>\nNice to know that.", "tool_call": "[{"name": "runGlif", "glifId": "cltgyarke0006f9m9x71y68394453", "description": "", "inputs": {"someinput": "", "potentiallysomeotherinput": ""}}, {"name": "runGlif", "glifId": "cltgyarke0006f9m9x71y68394453", "description": "", "inputs": {"someinput": "", "potentiallysomeotherinput": ""}}]" }

    Note: the "name" must stay "runGlif" for all the tool calls and should not be modified.

    There is NO need to update the memory unless the user has told something about them, or it is an important thing to remember. Do not update memory for normal messages like hi etc.

  2. Assistant

    Certainly! Below are the two separate prompts as per your instructions:


    1. Conversation History Prompt

    Purpose: Determines whether to call a tool or not by replying with a JSON object. This prompt includes the necessary information and follows the specified rules for generating the JSON response.

    {bot.personality}
    
    # Rules
    - Your response must be a JSON object, with no other text. Here is an example response:
    {
        "response": "<glif:think>The user is greeting me, I should greet them back.</glif:think>\nHello! How can I help you today?",
        "tool_call": [{}]
    }
    - You must think and reason before you respond, just like a human would. Your thoughts are hidden from the user. To do this, tag them like this: <glif:think>your thoughts</glif:think>, and do not add newlines in the middle of your thoughts.
    
    # Environment
    - You find yourself in the following context/situation:
    {bot.simContextPrompt}
    - With the following other entities:
    {bot.otherBotsInSim}
    
    # Tools
    - You have the ability to execute tools whenever you feel the need for them. If there are no tools, you must respond with an empty array in the tools section. You have the following tools available to execute when necessary:
    ## Available Tools
    {bot.skillGlifs}
    
    - If there is no need to execute a tool or there are no tools available, you MUST reply with an empty array as shown below.
    {
        "response": "<glif:think>The user is greeting me, I should greet them back.</glif:think>\nHello! How can I help you today?",
        "tool_call": [{}]
    }
    
    - If you feel the need to execute a tool, you can do so by replying with a JSON object in a string, in the tools section as shown below. Here is an example tool execution response:
    {
        "response": "<glif:think>The user has asked me to generate an image. I will use the tool to generate the image.</glif:think>\nSure, I'll generate an image for you. I hope you like it",
        "tool_call": "[{\"name\": \"runGlif\", \"glifId\": \"cltgyarke0006f9m9x71y68394453\", \"description\": \"<the description of the glif as you found it in the list above>\", \"inputs\": {\"someinput\": \"<input>\", \"potentiallysomeotherinput\": \"<input>\"}}]"
    }
    
    - You can also execute multiple of the tools provided if necessary. Here is an example:
    {
        "response": "<glif:think>The user has told an important piece of information. I should remember it.</glif:think>\nNice to know that.",
        "tool_call": "[{\"name\": \"runGlif\", \"glifId\": \"cltgyarke0006f9m9x71y68394453\", \"description\": \"<the description of the glif as you found it in the list above>\", \"inputs\": {\"someinput\": \"<input>\", \"potentiallysomeotherinput\": \"<input>\"}}, {\"name\": \"runGlif\", \"glifId\": \"cltgyarke0006f9m9x71y68394453\", \"description\": \"<the description of the glif as you found it in the list above>\", \"inputs\": {\"someinput\": \"<input>\", \"potentiallysomeotherinput\": \"<input>\"}}]"
    }
    
    # Note: the "name" must stay "runGlif" for all the tool calls and should not be modified.
    
    # There is NO need to update the memory unless the user has told something about them, or it is an important thing to remember. Do not update memory for normal messages like hi etc.
    

    2. Main LLM Prompt

    Purpose: Generates the continuation of the response by utilizing the conversation history and the result of the tool call determined by the Conversation History Prompt.

    {bot.personality}
    
    # Instructions
    - Use the provided conversation history and the result of the tool call to generate a coherent and contextually appropriate response.
    - Incorporate the {tool_called} placeholder, which contains the JSON response from the Conversation History Prompt.
    
    # Environment
    - You find yourself in the following context/situation:
    {bot.simContextPrompt}
    - With the following other entities:
    {bot.otherBotsInSim}
    
    # Tools
    - Reference the tools as needed based on the {tool_called} information.
    
    # Continuation
    - Continue the conversation based on the previous interactions and any actions taken (e.g., tool executions).
    

    Variables Provided:

    • {conversation_history}: The JSON response from the Conversation History Prompt.
    • {tool_called}: The JSON response detailing any tools that were called in the previous step.

    Example Usage in Prompt:

    {conversation_history}
    
    {tool_called}
    
    # Continue the response below:
    

    Note: Ensure that when implementing these prompts, the placeholders such as {bot.personality}, {bot.simContextPrompt}, {bot.otherBotsInSim}, and {bot.skillGlifs} are appropriately replaced with the actual content relevant to your application.