LLM paper reviews and auto-rebuttals, served with a warning

What it does

ChatReviewer feeds academic PDFs to an LLM—GPT-4o-mini on the web version, ChatGPT-3.5 locally—to produce quick summaries of strengths, weaknesses, and improvement suggestions. A companion script, ChatResponse, reads reviewer comments and generates point-by-point rebuttals. Both are available as local Python scripts or via Hugging Face Spaces.

The interesting bit

The author is openly paranoid about misuse: generated text is deliberately salted with mid-sentence warnings like “Generated by ChatGPT, no copying allowed!” to sabotage copy-paste peer review. It is a rare case where readability is sacrificed for ethics by design.

Key highlights

• Processes single papers or entire folders of PDFs locally
• Two-in-one repo: paper reviewer (chat_reviewer.py) and rebuttal generator (chat_response.py)
• Docker image bundles both services on separate ports
• Prompt logic asks the model which sections it wants to see before sending the full text
• Web UI hosted on Hugging Face Spaces; local use requires an OpenAI API key and VPN from China

Caveats

• Output is watermarked with ethics warnings that break reading flow, by the author’s own admission
• Local deployment requires VPN and an OpenAI API key; the Docker image must run on a non-China server
• Code is a modification of the earlier ChatPaper project, not a ground-up build

Verdict

Grad students and researchers who want a fast, second-opinion sanity check on drafts or reviewer letters will find it handy. If you need rigorous, domain-expert critique or are offended by inline capitalized warnings, look elsewhere.