Paper summary
Paper summary: AI Can Learn Scientific Taste
Reinforcement Learning from Community Feedback trains a "Scientific Judge" on signals like citations, and a "Scientific Thinker" that proposes research ideas; the judge generalises to future years and unseen fields.
- Community feedback, such as citations, is used as the reward for learning to judge and propose research ideas.
- The authors acknowledge citations are an imperfect signal; some influential work is cited late.