Tag
In the arXiv preprint (2606.25996), Meta's team proposes the Autodata method, where an AI agent acts as a data scientist to iteratively generate high-quality synthetic training and evaluation data. The core mechanism, Agentic Self-Instruct, forms a closed loop using an orchestrator, challenger, weak/strong solver, and judge, and introduces meta-optimization to evolve prompts, significantly improving data quality.
This paper proposes formalizing scientific discovery as a meta-optimization problem where LLMs generate and aggregate objective functions via correlation-weighted voting, applied to 3-SAT algorithm discovery using digital MemComputing, achieving a 67x speedup on large instances.
Autodata introduces an agentic data scientist that iteratively generates and refines synthetic training data, with meta-optimization to further improve data quality, achieving better results on computer science and legal reasoning tasks.
Autodata is a method that enables AI agents to act as data scientists to create high-quality synthetic training data through meta-optimization, achieving improved performance across computer science, legal reasoning, and mathematical tasks.
MetaEvo proposes a two-stage framework for continual evolution of LLM-based agents, using preference-based optimization to enhance principle abstraction and modular architecture for experience reuse, outperforming strong baselines on reasoning benchmarks.