AI system for multimodal theorem explanation via video generation
Top 30.4% on sourcepulse
This project provides TheoremExplainAgent (TEA), an AI system designed to generate long-form Manim videos that visually explain mathematical theorems. It targets researchers and educators seeking to deepen LLM understanding of theorems and identify reasoning flaws, offering a multimodal approach to complex concepts.
How It Works
TEA leverages large language models (LLMs) to generate detailed plans for Manim animations, which are then rendered into videos. It supports various LLM providers via LiteLLM and includes an optional Retrieval Augmented Generation (RAG) component for incorporating custom documentation. The system also features an evaluation module that uses LLMs to assess the quality of generated videos.
Quick Start & Requirements
conda create --name tea python=3.12.8
), activate it (conda activate tea
), and install dependencies (pip install -r requirements.txt
).portaudio19-dev
, libsdl-pango-dev
on Ubuntu), and LLM API keys (configured via .env
file).PYTHONPATH=$(pwd):$PYTHONPATH
.Highlighted Details
Maintenance & Community
The project is associated with TIGER-AI-Lab and has released code and evaluation components. Further community interaction details are not explicitly provided in the README.
Licensing & Compatibility
Limitations & Caveats
Evaluation requires specific LLMs (Gemini, GPT-4o) and video/subtitle files. The disclaimer advises against commercial use, and the project is presented "as is" without warranties.
1 week ago
1 day