This year for STOC, we decided to run an experiment to explore the use of Large Language Models in the theoretical computer science community, and we’re inviting the entire community to participate.

We—a team from the STOC PC—are offering authors the chance to get automated pre-submission feedback from an advanced, Gemini-based LLM tool that’s been optimized for checking mathematical rigor. The goal is simple: to provide constructive suggestions and, potentially, help find technical mistakes before the paper goes to the PC. Some important points:

This is 100% optional and opt-in.
The reviews generated WILL NOT be passed on to the PC. They are for your eyes only.
Data Privacy is Our #1 Commitment. We commit that your submitted paper will NOT be logged, stored, or used for training.
Please do not publicly share these reviews without contacting the organizing team first.

This tool is specifically optimized for checking a paper’s mathematical rigor. It’s a hopefully useful way to check the correctness of your arguments. Note that sometimes it does not possess external, area-specific knowledge (like “folklore” results). This means it may flag sections that rely on unstated assumptions, or it might find simple omissions or typos.

Nevertheless, we hope you’ll find this feedback valuable for improving the paper’s overall clarity and completeness.

If you’re submitting to STOC, we encourage you to opt-in. You’ll get (we hope) useful feedback, and you’ll be providing invaluable data as we assess this tool for future theory conferences.

The deadline to opt-in on the HotCRP submission form is November 1 (5pm EST).

You can read the full “Terms of Participation” (including all privacy and confidentiality details) at the link below.

This experiment is being run by PC members David Woodruff (CMU) and Rajesh Jayaram (Google), as well as Vincent Cohen-Addad (Google) and Jon Schneider (Google).

We’re excited to offer this resource to the community.

Please see the STOC Call for Papers here and specific details on the experiment here.