According to Ethan Mollick on X, a study shows a small model trained on citation signals can predict which research papers will become high-impact hits, indicating AI can learn judgment about quality beyond execution; as reported by Ethan Mollick, social signals like citations, upvotes, and shares provide supervisory signals that encode community taste and future impact. According to the linked paper (via Ethan Mollick’s post), training on historical citation trajectories enables forecasting of future citations, suggesting practical applications for venture scouting, R&D portfolio management, and editorial triage in academia and industry.