Autodata: Using AI agents as data scientists to generate high-quality synthetic training data

Summary

This paper introduces Autodata, a method that uses AI agents as data scientists to create high-quality synthetic training and evaluation data. The approach involves training (meta-optimizing) a data scientist agent that learns to produce increasingly better data. The paper describes a practical implementation called Agentic Self-Instruct, and presents experiments across computer science research, legal reasoning, and mathematical reasoning tasks. Results show improved performance compared to classical synthetic data creation methods, with further gains from meta-optimizing the agent itself. The authors argue this direction could fundamentally change how AI training data is built.

Source

Twitter / XAutodata: Using AI agents as data scientists to generate high-quality synthetic training dataarxiv.org