Simiverse: Simulation Framework

Overview

Simiverse is a versatile simulation framework designed to be a universal plugin for applications, providing the necessary base simulation class and a variety of simulation modules. It allows for easy integration into different systems and can be utilized across various domains such as training data generation for machine learning models, evaluating AI alignment, simulating complex decision-making processes, and more. The framework provides tools for managing entities, environments, and decision-making within simulations, with the flexibility to use AI-driven decision models through an extensible client architecture.

Simiverse consists of a base class, Simulation, which supports both traditional and AI-assisted decision-making workflows. Additionally, the framework includes specialized modules, like the LLMClient and OpenAIClient, to support AI interactions for more advanced simulations. This repo is integrated with a web platform designed to run simulations efficiently and interactively.

Table of Contents

1. Prerequisites
2. Features
3. Simulations
4. Setup Instructions
5. How to Run
6. Early Thoughts

Prerequisites

• Node.js installed on your machine.
• OpenAI API key for AI-assisted decision-making.
• Git for cloning the repository.

Features

• Universal Simulation Framework: Build your simulations using the base Simulation class, with support for integrating custom modules.
• Dynamic Entity Generation: Simulate diverse entities with customizable attributes.
• AI-Assisted Decision-Making: Easily integrate AI models for decision-making processes, using the LLMClient and OpenAIClient modules.
• Custom Simulation Modules: Extend the framework by adding specific modules to fit various use cases such as training, evaluation, and AI alignment.
• Web Platform Integration: Seamlessly run and manage simulations on a web platform built for simulation execution and interaction.

Simulations

Survival Simulation

Description: The Survival Simulation models global and national decision-making in a resource-constrained world. It demonstrates the delicate balance between cooperation and defection, where nations must decide whether to collaborate or claim resources to survive. The simulation tracks global resource levels and the collapse of nations based on their decisions.

• Simulation Type: Resource management, decision-making, global collapse.
• Primary Focus: Cooperation vs. defection dynamics among nations.
• Features:
  • Dynamic nation generation with unique resource states.
  • Resource depletion and its impact on global stability.
  • AI-assisted decision-making for nations' actions.

Example Usage:

import { SurvivalSimulation } from "simiverse/simulations";
import { OpenAIClient } from "../llms/openai";

const simulation = new SurvivalSimulation(entities, environment, {
  steps: 10,
  type: "sim",
  openaiApiKey: "your-api-key",
  onStepComplete: (eventData) => console.log(eventData),
});
simulation.run();

Setup Instructions

1. Clone Repository:

   git clone https://github.com/dev-diaries41/simiverse.git
   cd simiverse

2. Install Dependencies:

   npm install

3. Configure Environment:

   • Ensure you have an OpenAI API key.
   • Create a .env file in the project root and add your API key:

     OPENAI_API_KEY=your_api_key_here

How to Run

1. Run the Simulation:

   npm start

Early Thoughts

Here are some early ideas and potential features I may implement into the Simiverse framework in the future:

• 3D Avatar Generation: Allow users to generate 3D avatars simply by uploading an image. These avatars could be used within simulations to represent entities or characters in the environment.

• Brain-Computer Interface (BCI) Integration: Enable simulations to interact with brain-computer interfaces for real-time feedback or decision-making based on neural activity.

• Virtual Reality (VR) Integration: Provide VR support for immersive simulation experiences. Users could navigate or participate in simulations in real-time using VR headsets, creating an interactive environment.

• Hardware Integrations: Explore possibilities to integrate the framework with various hardware platforms. This could include using sensors or IoT devices for real-time data collection, feedback, or control within simulations.

• 3D Holographic Displays: Experiment with holographic displays for visualization of simulation data. This would provide a more immersive way to interact with and analyze simulation outcomes, potentially including multi-dimensional data.

• Using Simulation Data to Train ML Models: Leverage the data generated from simulations as training material for machine learning models. This would allow the creation of robust models by simulating a wide range of conditions and environments, which could be used for reinforcement learning or other AI applications.

These ideas are in the early stages and will evolve as the framework continues to develop.