Recursa: Toward Recursively Self-Improving Foundation Model Systems

What happened: Recursa is a self-improving foundation model system that adapts to new domains by learning from its own experience, storing skills and reflections in a knowledge repository and validating improvements without human supervision or stronger teachers. Key numbers: On Harvey LAB's Firm Knowledge benchmark, Recursa raised the all-pass rate on held-out test Why it matters: The system targets domains that lack large expert-created training data or reliable supervisory signals, using parallel learners to propose improvements that are validated and merged back into the repository.

Recursa is a self-improving foundation model system that adapts to new domains by learning from its own experience, storing skills and reflections in a knowledge repository and validating improvements without human supervision or stronger teachers.

The system targets domains that lack large expert-created training data or reliable supervisory signals, using parallel learners to propose improvements that are validated and merged back into the repository.