Hugging Face has highlighted a new research paper introducing ScienceBuddy, an interactive workspace designed to embed continually improving scientific agents into researchers' everyday workflows. According to the paper, the system assists researchers with scientific tasks while simultaneously converting their requests, feedback, and execution evidence into tasks and evaluation rubrics for continual learning.
At the heart of the system is a paradigm the authors call recursive-in-recursive self-improvement, which couples harness evolution with model reinforcement learning. In the inner recursion, the harness is improved while the model is held fixed; in the outer recursion, the model is trained under the improved harness. The paper describes this as a mutually reinforcing loop: harness evolution shapes the training experience, while model learning opens new opportunities for harness adaptation.
The authors present case studies covering researcher interaction, harness refinement, and model learning, with benchmark cases spanning four scientific task families. The work is framed as a step toward what the paper terms "discovery intelligence" — scientific AI that advances through sustained collaboration with researchers and evolves alongside the research it supports.
ScienceBuddy is being released as a research product, with a project website at science-buddy.io and code available on GitHub. The paper appears on Hugging Face's papers page and currently has no associated collections.