Master the AI Conference Rebuttal When Reviews Exceed Your Space

Navigating the character limit in rebuttals can be challenging, especially when faced with extensive reviewer comments.

3 min readMachine Learning

The reviewer dumps forty bullet points into your rebuttal form, and you have 2,500 characters to answer them. This is not a failure of your writing. It is a failure of the process. When reviewers outsource their thinking to an LLM, the output is a firehose of surface-level complaints, formatting nits, missing citations for obvious theorems, rewordings that read like an intern paraphrasing your abstract. None of it is malicious. All of it is useless. The system was not designed to handle a reviewer who generates a page of commentary before forming a single opinion, and you are the one who gets squeezed.

The practical step is brutal but honest: do not try to answer every point. Reviewers who paste LLM output are not reading line by line. They scanned, they pasted, they submitted. Your rebuttal should match that energy. Pick the three or four criticisms that actually touch the core of your contribution, the novel method, the surprising result, the clean ablations, and answer those with evidence. Let the rest stand unanswered. If the reviewer complains that you ignored their formatting concern, the meta-reviewer will see that you spent your characters on substance. This is not rudeness. It is triage.

The deeper problem is that the review process is being quietly reshaped by the tools meant to assist it. Reviewers are not your enemies. They are overloaded, chasing deadlines, and reaching for whatever promises speed. An LLM promises that, but it trades thoroughness for volume. You cannot fix that in a 2,500-character box. What you can do is force the conversation back to the work. State plainly: "We focused on the claims about generalizability because that is the central contribution." Then show your evidence. The reviewer, and the area chair, will recognize the shape of a real argument.

Treat the character limit as an editorial constraint, not a negotiation. Write a draft, cut it to half, then cut again. Every sentence should either correct a specific misunderstanding or strengthen a claim the reviewer already accepted. If a sentence does neither, delete it. You are not writing a letter. You are packaging a decision aid for the area chair. Make it dense, make it honest, and do not apologize for the space you did not have.

From Machine Learning

First time submitting to an AI conference here.

How do you deal with the situation where a reviewer gives you a long list of review comments, but you only have a 2.5k character limit in your rebuttal?

Read the original at Machine Learning