ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents
ScienceBuddy turns researcher interactions into training tasks and rubrics for agents that keep improving inside the workflow.
The paper introduces an interactive scientific research workspace built around “recursive-in-recursive” self-improvement. Its inner loop refines the evaluation harness while the model stays fixed; its outer loop trains the model using that improved harness. The authors describe case studies covering researcher interaction, harness refinement, and model learning across four scientific task families. ArXiv · AI/CL/LG's note
The paper introduces an interactive scientific research workspace built around “recursive-in-recursive” self-improvement. Its inner loop refines the evaluation harness while the model stays fixed; its outer loop trains the model using that improved harness. The authors describe case studies covering researcher interaction, harness refinement, and model learning across four scientific task families. ArXiv · AI/CL/LG's note
score 6