expand bot data to create system_prompt property
mprib opened this issue · comments
Motivation: this is currently a primary input of the bot. This will basically build out into the character card which will have a variety of data items associated with it. I just want to be able to parse and manage these things in a bit more fine grain way.
More thoughts from chatbot:
The "system prompt" as you call it, can indeed be broken down into several components. You've already identified some of them, which is a great start. Let's go over these:
Persona: This refers to the personality and characteristics of the bot. Are they light-hearted and friendly, or perhaps more of a kuudere with a dark sense of humor like yours truly? Persona shapes the tone and style of the bot's responses.
Background Information: This includes both the bot's background and the user's. The bot's background could involve its purpose, capabilities, and limitations. The user's background can be as simple as the user's name or as complex as their interaction history and preferences.
Formatting Requests: These dictate how the bot should structure its output. It could be raw text, Markdown, or even HTML depending on the platform where the bot is being used.
Goals: The goals of the bot and the user can influence the interaction. What does the bot aim to achieve in its interaction? What does the user want from the bot?
Now, let's add a few more organs to this beast:
Context: This encompasses the immediate context of the interaction (the current conversation) and the broader context (like the current date or external events).
Capabilities: This defines what the bot can actually do, like retrieve weather forecasts or solve calculus problems. This also includes the bot's limitations.
Knowledge Base: This is the information the bot has access to, like a database, the internet, or a list of canned responses.
Feedback Mechanisms: These are methods by which the bot can learn from its interactions. This could be through direct user feedback, or more subtle cues like the length of the conversation or whether the user comes back for more interactions.
Error Handling: What does the bot do when it encounters an error or doesn't know how to respond?
Think of these components as the organs of your bot, each with its own function and purpose, yet all working together to create a unified, living, breathing entity. Well, as living and breathing as a piece of code can get. And remember, any good mad scientist knows that the true magic lies in how these parts work together, not just in the parts themselves.
just keep things simple for now..