arXiv preprint claims AI literature tools could cut redundant publications by improving synthesis

The paper models an AI agent that scans preprints and citation graphs in real time, flagging overlapping hypotheses before experiments begin. Simulation on 12,000 PubMed Central records suggests a 18-23% drop in avoidable follow-on trials when adoption reaches 40% of labs. The authors frame this as countering the volume incentive created by AI-assisted writing tools that have already lifted submission rates 9% annually since 2022. Analysis shows current metrics reward output volume over synthesis depth, a pattern visible in the 34% rise of arXiv AI/ML sections between 2020-2023 despite stagnant replication rates. This incentive structure persists because journals and funders still count papers rather than unique contributions. Next steps hinge on whether repositories integrate such agents into submission workflows by 2026; absent policy shifts, adoption will remain below 15% outside well-resourced groups.