Meet Reviewer 2 before Reviewer 2 meets you.
Three harsh reviews in your venue's exact form. Five minutes. Free.
First review free, no card. Your PDF stays private.
Drop in your PDF. Get three conference reviews, Reviewer 2 included.
- Reading the paper
- Three reviewers are reading the paper
- Area chair is writing the meta-review
Clearly written, and the ablations are sensible.
The gains are within noise: three seeds, no confidence intervals. The novelty claim ignores mixture-of-LoRA work.
A small gain over LoRA for real added complexity.
Hand the fix list to Claude Code.
Clearly written, and the ablations are sensible.
The gains are within noise: three seeds, no confidence intervals. The novelty claim ignores mixture-of-LoRA work.
A small gain over LoRA for real added complexity.
I'm revising “Sparse Routing Adapters” for NeurIPS 2026. Help me fix these in the paper:
- Report variance over 5+ seeds
- Compare against mixture-of-LoRA baselines
- Report routing latency
I'll work through the three fixes in order, starting with the seeds.
Upload the new version. See what moved.
Clearly written, and the ablations are sensible.
The gains are within noise: three seeds, no confidence intervals. The novelty claim ignores mixture-of-LoRA work.
A small gain over LoRA for real added complexity.
- Reading the paper
- Three reviewers are reading the paper
- Checking what changed since v1
Since your last version: 2 fixed · 1 partly
Report variance over 5+ seedsTable 1 reports mean ± 95% CI over 5 seeds.
Compare against mixture-of-LoRA baselinesTable 2 adds MoLoRA and a dense adapter.
Report routing latencyBatch size 1 only; larger batches are still missing.
Predicted scores. Treat them as a stress test, not a forecast.
Review forms for 29 venues: NeurIPS, ICLR, ICML, AISTATS, UAI, TMLR and 23 more
Start with a free review
Then unlock the paper you’re submitting for $9 once, or get Pro for $19 a month if you have several.
- Free
- $0
- first review
- Unlock
- $9
- one paper, once
- Pro
- $19
- every paper / month
Questions
Privacy, accuracy, and what this is not.
Who can see my paper?+
Only you. Your PDF sits in a private bucket, is served only to your signed-in account and is never used to train models. Uploading a paper under double-blind review reveals nothing to reviewers, venues or other users, and venue LLM rules are aimed at reviewers, not authors checking their own work (our LLM policies guide lists each venue's rule). Delete a paper and its file, reviews and chat are gone.
How accurate are the scores?+
Treat them as a stress test, not a forecast. Real reviewers disagree a lot, and ours can miss what a domain expert would catch or flag issues that are not real. The value is in the specific, cited criticisms you can fix or answer.
My paper is already submitted. Is this still useful?+
Yes. Reviews take weeks, so seeing the likely objections now gives you time to run the missing experiment before the author response opens. When the real reviews land, import them and get a draft rebuttal. If it is rejected, run the revision against the next venue's form.
What's free, and what does it cost?+
Your first review is free on Claude Sonnet 5.5: three reviews, a meta-review, a predicted score and the top 3 fixes. Unlocking that paper ($9 once) adds the full fix list and checklist, unlimited re-runs on Claude Opus 5.5, revision checks and rebuttal drafts. Pro ($19/month) unlocks every paper.
Are you affiliated with NeurIPS, ICLR, ACL or any other venue?+
No. reviewpaper.ai is independent. We model each venue's public reviewer form and guidelines, and each venue page links to the official source.
Can I upload a paper I was assigned to review?+
reviewpaper.ai is built for authors checking their own papers. Most venues forbid reviewers from putting a confidential submission into any external LLM tool, and some desk-reject the reviewer's own papers for it. Check your venue's rule in our LLM policies guide first.
Hear it from Reviewer 2 while you can still fix it.
- First review free
- Private by default
- Every point cited