Agentic Data Synthesis
Methods for generating and evolving synthetic training data with agentic pipelines, driven by learner or verifier feedback.
4 papers
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.
How this thread developed
Oct 2026 · University of Washington, Carnegie Mellon University
Recursive synthesis turns unsolved problems into useful reinforcement learning data
Converts unsolved problems into RL data with a useful mix of successes and failures for verifiable-reward training.
2 further papers
Oct 2026 · University of Science and Technology of China, State Key Laboratory of Cognitive Intelligence
Self-improving data synthesis adapts embedding models to new tasks with minimal human input
Iteratively improves synthesis instructions using feedback on trial outputs to generate training data for embedding models.
2 of 4 papers shown