CodeScientist

CodeScientist is a semi-automated scientific discovery system that can design, iterate, and analyze scientific experiments expressed in Python code. It uses large language models (LLM) as mutators to perform genetic mutations on scientific articles and code examples, thereby generating new experimental ideas. These experimental ideas can be automatically created, run, and debugged within containers by the experiment builder. Upon completion, CodeScientist generates a report on the results.

The use cases of CodeScientist include:

• Automated Scientific Experiment Design: Used to automatically generate experimental ideas in specific fields and convert them into executable code.
• Experiment Iteration and Analysis: Used for semi-automated iteration of experiments and analysis of experimental results, generating reports.
• Field Exploration and Discovery: Helps researchers explore new research directions and discover potential scientific breakthroughs.
• Education and Learning: Used for teaching and learning the design and analysis of scientific experiments.
• Benchmark Testing and Evaluation: Used to compare different experimental methods and evaluate their performance.

CodeScientist mainly has two operating modes:

• Human-Machine Collaboration Mode: Requires human participation to help build code examples, screen experimental ideas, and provide comments that aid in implementation.
• Fully Automatic Mode: Can run fully automatically, but the efficiency of producing scientific results is lower.