LLM Stress Tests: Pushing the Boundaries of AI Code Generation
This repository is dedicated to exploring and stress-testing the capabilities of Large Language Models (LLMs) in code generation, particularly focusing on what they can achieve with minimal, concise, and sometimes ambiguous input.
Mission
The core idea is to provide an LLM with a high-level concept or a set of core feature requests and observe its ability to:
- Translate abstract requirements into functional code.
- Select appropriate libraries and technologies (when not specified).
- Implement complex logic and interactions.
- Structure code for readability and maintainability (to varying degrees).
- Handle implicit requirements and make reasonable assumptions.
- Generate complete, runnable applications within given constraints (e.g., a single file).
Why Stress Test?
- Understand Capabilities: To gain a deeper understanding of current LLM strengths and weaknesses in practical software development scenarios.
- Identify Limits: To find the "breaking points" where LLMs struggle or fail, providing insights into areas needing further research and development.
- Improve Prompt Engineering: To learn how to formulate prompts more effectively to guide LLMs towards desired outcomes, especially for complex tasks.
- Showcase Potential: To demonstrate the impressive potential of LLMs as tools for rapid prototyping, learning, and even production-level assistance.
Live Demo
A live, unified demo of all projects is available via GitHub Pages:
Projects
| Project | LLM | Tags | Preview |
|---|---|---|---|
| 3D Village Builder | Gemini-2.5Pro | 3D, Game, Three.js |  |
| Wealth Visualizer | Gemini-2.5Pro | 3D, Data-Viz, Three.js | |
| Music Synthesizer | Horizon-Beta | Audio, Music, WASM | |
| Time-Travel Dashboard | Horizon-Beta | 3D, Data-Viz, WebGL | |
| Poetry Generator | Horizon-Beta | Poetry, Text-Gen | |
| Solidity Generator | Horizon-Beta | Solidity, Code-Gen, Legal-Tech |
How to Interpret Results
The projects here are not necessarily examples of "best practices" in software engineering if a human were solely responsible. Instead, they are a reflection of the LLM's interpretation and execution based on the input provided. Manual clean-up or refinements by a human developer might be noted, but the initial output from the LLM is the primary focus.
Contributing
While the primary goal is to document LLM-generated projects based on specific prompts, suggestions for new stress-test project ideas are welcome! Please open an issue to discuss your idea.
If you replicate a project or try a similar prompt with a different LLM, feel free to share your results and observations.
License
This project is licensed under the MIT License. See the LICENSE file for details.
Disclaimer
[!WARNING] The code generated by LLMs and hosted in this repository is for experimental and educational purposes. It may contain bugs, inefficiencies, or security vulnerabilities. Always review and test AI-generated code thoroughly before using it in critical applications.