AI Slop Papers Make Reviewing Easier: Only 25-50% Deserve Careful Reads

tallinzen · x · 2026-09-17

tallinzen observes that AI slop papers have made ML conference reviewing easier on average, since many submissions quickly reveal they don't merit careful review — reviewers really only need to read 25-50% of their batch. The quoted context: Nihar Shah ran a heroic TMLR experiment, spending 20-25 hours over two weeks interviewing authors of seemingly low-quality submissions, confirming that these authors have no idea what is going on in their own papers.

Original post →

More from Research

Research channel →