NeurIPS

When peer review goes quiet, trust in the process fades.

A reviewer's job doesn't end when the rebuttal lands, but it seems many participants on both sides of the NeurIPS aisle have simply stopped engaging.

4 min readMachine Learning

A single anonymous post about a quiet review cycle might normally fade into the noise of conference season. But when a reviewer describes a batch where two of four papers went completely dark, one withdrew, and only a single author bothered to rebut, that stops being anecdotal. It becomes a signal about how we treat the social contract of science. The original poster is right to be puzzled, and their observation deserves more than a shrug.

Let's be direct about what is happening. The review period is not just slow; it is revealing a structural shift in incentives. The poster notes a pattern of reviewers disappearing after initial reviews, and authors either withdrawing or simply ghosting. The gambling analogy they raise is uncomfortable but apt. If submitting papers becomes a low-stakes lottery ticket, then the act of shepherding a submission through rebuttal, revision, and response becomes a cost that rational actors avoid. Why invest time defending a borderline score when the expected value of engagement is near zero? This is not a failure of individual will. It is a market failure in academic reputation, where the currency of responsiveness has been devalued. We have all seen the rise of multi-submission strategies, but the quiet casualty is the review process itself, which relies on the assumption that both sides show up.

This connects directly to a broader problem we have covered in our Clean Data Starts With Catching AI Slop Before It Skews Your Model, where noise and low-quality signals corrupt an otherwise functional system. Just as AI detectors struggle to separate genuine feedback from generated filler, our peer-review system is now struggling to separate genuine engagement from performative submission. When authors go silent, they are not just harming their own papers. They are eroding the calibration of every reviewer who invested time reading their work, and by extension, the quality of decisions for everyone else in the pool. The reviewer in the post who stayed active despite withdrawing their own paper is the exception, but exceptions do not sustain a community.

What would we tell a reader who asked about this? Stop treating the rebuttal as optional. If you submit a paper, you owe the reviewers a response, even if that response is a withdrawal. Silence is not a neutral act; it is a decision that tells reviewers their labor has no value. The poster's frustration is legitimate, and the fact that they were the only reviewer to respond to a rebuttal is a damning indictment of the current culture. We would say this: the next time you are tempted to let a paper sit without a rebuttal, remember that you are not just leaving a review thread empty. You are teaching reviewers that their time is worthless. That lesson sticks.

The concrete point to watch is whether conference organizers start penalizing ghosting behavior, perhaps by implementing submission caps or requiring a minimum engagement score. If they do not, the dead review period will become the new normal, and we will all be poorer for it. But the change does not have to start at the top. It starts with a single reviewer, like the one who posted this, choosing to stay active and model the behavior they want to see. That is the only reliable signal we have left.

From Machine Learning

I’ve seen a lot of people whose reviewers went silent after initial reviews, but I am also noting abnormally quiet authors. I ultimately withdrew my paper, but stayed an active reviewer. Out of my batch of 4 papers, one withdrew, one posted a rebuttal, and two have been completely silent. Of the two papers with radio silence, I think one had borderline scores. I was also the only reviewer who responded to the one paper with a rebuttal.

Read the original at Machine Learning