A recent paper investigates the composition of the NeurIPS and ICML Position Paper Tracks for 2025. The authors argue that the publicly accessible submissions are dominated by reformist critiques rather than direction-setting work. An audit of the submissions found that three-quarters of them critique existing benchmarks, evaluations, or methodologies, but do not necessarily provide new evidence to support their claims. This is in contrast to widely recognized agenda-shifting papers, which often introduced new tools, protocols, or methods to advance the field. --- Why it matters: This study matters because it highlights a potential issue with the way research is being presented and evaluated in top AI conferences. The findings suggest that the current format may not be conducive to advancing the field, and that changes are needed to encourage more innovative and impactful submissions. Source: https://arxiv.org/abs/2608.16894