Post

I’d also add that reviewers shouldn’t be biased against for NOT using LLMs, e.g. “This problem is trivial and uninteresting because the free version of <model> solved it instantly” should not be the sole reason for rejection.

In general, the cost/time/number of tokens to solve a problem shouldn’t be used to measure how interesting a problem is

This needs to be qualified a bit. I agree that we shouldn’t measure individual papers by such a metric. But AI will change what problems are considered difficult and worth publishing at a broader level. The PDE community had a similar change with numerical methods becoming affordable.

Obviously recommendation 4 (that papers must be submitted to arXiv) caught my eye. Two concerns: (a) arXiv moderators sometimes reject a submission and ask authors to go through peer review before reconsideration. This is usually done for major problems or poor writing. But it could place authors

in a Catch-22. In effect, they won't be able to submit to any other conference that also requires ArXiv submission. (b) arXiv will need more moderators in TCS-related categories to handle the increased load. (We need more in math.CO right now, as it is receiving 100s of AI-assisted submissions.)

To elaborate on Point 8: of course, this acknowledges uncertainty, which is scary (but unfortunately obvious). The good and uplifting part is that it's explicitly prescribing that there should be a community focus (with funding!) on junior researchers, as opposed to "doom and free for all."

Item 7 is also crucial—I am very glad to see it make explicit. "This changing landscape is also obfuscating what early career researchers view as success and how their mentors should guide them. In particular, AI risks accelerating an “arms race” in the number of research publications."