Skip to content

GATO-Framework/shappie

Repository files navigation

Sandbox for Heuristic Assessment: a Potentially Prolific Interaction Experiment

Shappie is your conversational companion, built on the power of Large Language Models (LLM) such as GPT-4.

Key Features

Engaging with Shappie

@Shappie - That's all it takes to spark a conversation with Shappie. Reply directly to Shappie's messages, or mention the bot in your message to trigger a response. Expect Shappie to reply in the same channel, contributing to the conversation seamlessly.

Note: Shappie strictly adheres to the rules of engagement, which means it won't respond in unauthorized channels. If you call upon Shappie where it's not allowed, it will guide you to a place where you can converse freely.

Invoke Shappie's Tools

Shappie isn't just an engaging conversationalist, but also a versatile toolbox. It listens for keywords in your messages and, if detected, will leverage the appropriate tool to assist you. Here are some examples:

  • When to Meet: For scheduling a meeting, Shappie can share a link for when2meet.com helping to find a suitable time slot for everyone.
    • Example: “Lets set up a meeting to talk about that!”
  • Layer Information: Ask Shappie about any GATO layer, or request a summary.
    • Example: “Can you tell me about layer 2 of GATO?”
  • Paper Search: Hunting for papers on a specific topic? Let Shappie search arXiv for you.
    • Example: “Can anyone recommend a good paper about logical neural networks?”
  • Fallacy Detection: Drop a 💩 emoji as a reaction to any message to have Shappie identify any logical fallacies it contains.

Remembering with Shappie

Short-Term Memory

For the immediate conversation at hand, Shappie remembers the last ten messages in the channel by default.

Long-Term Memory

Every message in channels Shappie has access to is saved, irrespective of whether Shappie is permitted to respond there. While you currently can't interact with these historical messages, they form a rich reservoir of data for future analytical efforts.

Releases

No releases published

Packages

No packages published