Navigating conference rebuttals shouldn't require a degree in character-count strategy. The situation posted by this researcher, finding UAI's 2,500-character limit only after finishing a rebuttal written for ICML's 5,000, is exactly the kind of friction that undermines good scholarship. The platform should serve the science, not the other way around.
The core question is practical: can a public comment serve as a continuation of the rebuttal? Our view is clear, use the tools as they were designed. Burying rebuttal content in a public comment and asking reviewers to follow a breadcrumb trail is risky. Area Chairs and Senior Area Chairs read the official rebuttal space first. Splitting your argument across two sections creates a burden on reviewers, and some may simply miss the continuation. Worse, the ambiguity the researcher describes, "is this ground for desk rejection?", is a sign that the system itself is unclear. When a conference fails to communicate its own limits clearly until after reviews are out, it has already failed its authors.
The best move here is honest compression. A 2,500-character limit forces authors to prioritize their strongest points. Lead with the most critical experiment, the tightest theoretical justification, or the most direct response to a misunderstanding. If a reviewer asked three questions, answer the one most likely to change their score. You cannot cover everything, but you can be decisive about what matters. Shortening the rebuttal is not giving up, it is showing that you understand what is at stake and that you respect the reviewer's time. If the author truly needs to extend beyond the limit, the cleanest option is to place the full rebuttal directly in the public comment section *instead* of the rebuttal field, with a note explaining the choice. That is a single destination for the reviewer to check, not a scavenger hunt.
This situation reveals a deeper problem with how conferences handle author experiences. When a researcher has to ask on Reddit whether using a platform's own features will get them desk-rejected, the tool has become an obstacle. AI-native systems should automate these constraints, or at least surface them early. A conference submission system that cannot warn a user about a character limit during drafting is not ready for the workflows researchers actually need. The industry must do better, not by adding features, but by removing friction. For now, compress your argument, put it where it belongs, and move on.