Tag
A new paper introduces GIANTS-4B, a 4-billion-parameter model trained with reinforcement learning to predict scientific insights by combining ideas from foundational papers, achieving higher similarity and citation potential than larger models like Gemini 3 Pro.