Interactive demo
Disruptiveness scoring
Seeded from this week's #1 paper. Drag factors to recompute the composite — same five-axis model used on cards and articles.
Calibrated to
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
arXiv:2501.07563 · editorial 87/100
Outcome-driven RL for reasoning at scale challenges the assumption that massive human CoT labels are required, reshaping how frontier labs train reasoning systems.
composite 87/100
Is this a new idea, proof, architecture, or measurement — or a small delta?
If true, how much do roadmaps, products, or theory change?
How active is this subfield right now on arXiv and in labs?
Can someone act on this soon — devices, code, trials — or is it pure theory?
Does it challenge orthodoxy or invite healthy debate? (Not drama for its own sake.)
Scores are triage, not peer review. See methodology · open this paper · curation demo.