SelfImprovingAgent is an AI-driven system that autonomously generates, tests, and refines Python code. The system uses language models to generate code based on a prompt, runs the code to check for correctness, and refines the code based on the feedback received from execution. Over multiple iterations, the agent "self-improves" its code, learning from its successes and failures.
- Code Generation: The agent can generate Python code based on user-provided prompts.
- Error Handling & Feedback: The agent evaluates the generated code by executing it and refines the code based on the feedback (success or errors).
- Iteration: The agent runs in multiple iterations, improving the generated code each time.
- Prompting: The agent starts with a base prompt (e.g., write a Python function to sort a list).
- Code Generation: The agent sends this prompt to a language model (e.g., GPT-based models).
- Evaluation: The generated code is written to a file and executed.
- Refinement: If the code executes successfully, it continues, otherwise, the agent refines the code based on error feedback.
- Iteration: This process repeats multiple times for incremental improvements.
The agent can be used with multiple LLMs (Language Models), such as:
- GPT-3
- Grok (if available)
- Custom Models (For any model that supports chat completions).
To change the model, simply set the model parameter when creating the SelfImprovingAgent object.
- Clone the repository:
git clone https://github.com/NullLabTests/SelfImprovingAgent.git cd SelfImprovingAgent
Install dependencies:
bash Copy Edit pip install openai Set your OpenAI API key:
bash Copy Edit export OPENAI_API_KEY="your-api-key" Run the agent:
bash Copy Edit python SelfImprover.py How to Evolve Further This agent could evolve in several ways:
Different Programming Tasks: Extend the agent's capabilities to generate solutions for other programming problems (e.g., web scraping, machine learning models). Meta-Learning: Implement meta-learning to allow the agent to adapt to new kinds of tasks without needing manual intervention. Model Switching: Allow the agent to automatically switch between different models depending on the task complexity or success rate. Relevant Papers Neural Architecture Search: A Survey Generative Models for Programming: A Study Contributing Feel free to fork and submit pull requests. Contributions are welcome!
