The idea of capping ML conference acceptances at 1,000 papers sounds like a clean solution to a messy problem, but it would likely trade one set of flaws for another. We should say plainly that a hard cap treats the symptom, overflowing programs, while ignoring the deeper issue of how we define and reward research contributions. As we explored in Navigating conference submission caps in an era of expanding research teams, the real pressure comes from teams scaling up output without a corresponding increase in venues or review capacity. A cap doesn't fix that; it just raises the stakes for every submission.
Proponents argue that scarcity would force researchers to submit only their strongest work, sharpening overall quality. There is some logic there. When every slot is precious, authors might self-filter more aggressively. But the machine learning community already knows that acceptance rates below twenty-five percent create a lottery effect, not a quality filter. A cap would make that worse. Reviewers already face fatigue with thousands of papers per cycle. Adding a hard numerical limit would turn the process into a competition over signaling and networking rather than scientific merit. The risk is that incremental but solid work gets squeezed out in favor of flashier results that may not replicate. Our piece on Exploring the Productivity Surge in China's Machine Learning Research showed how increased output from large research groups is already straining review norms. A cap would disproportionately affect those same groups, but not necessarily in a way that rewards depth over volume.
What would a better approach look like? The community could explore tiered acceptance categories, where some papers are presented in poster sessions and others in full talks, without an artificial ceiling on total accepted work. Another option is to expand the number of reputable venues and let the market of ideas sort itself out, rather than centralizing prestige into three conferences. The key is to reduce the bottleneck without making acceptance arbitrary. A hard cap of 1,000 would simply move the arbitrariness from the review threshold to the cutoff line.
The specific consequence to watch is how a cap would affect junior researchers. They are the ones who need conference exposure to build reputations and find collaborators. If a cap makes acceptance more competitive, senior researchers with established networks and bigger labs will dominate the limited slots. That outcome would contradict the field's stated values of inclusivity and broad participation. We should be asking not just how to limit papers, but how to expand the ecosystem so that good work finds an audience regardless of conference capacity.