Reproduce “Reinforcing Agentic Creativity in Scientific Ideation with Night Science” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Priyanka Kargupta, Silviu Cucerzan, Shweti Mahajan, Allen Herring, Jiawei Han, Ryen W. White, Sujay Kumar Jauhar
AI NIGHT-SCIENTIST trains language-model agents to choose actions and creativity levels for proposal generation, using reinforcement learning to reward precedence, feasibility, and relevance. It compares proposals with reconstructed references using pairwise judgments. For Qwen3-14B, Night-14B’s citation-impact win rate is 33.18% versus 1.15%, while originality is 68.68% versus 2.53%. These results concern proposal comparisons, not actual citations or validated advances.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.