Academic researcher Nihar Shah conducted a series of interviews with authors of low-quality submissions to the Transactions on Machine Learning Research (TMLR) journal over a two week period. Shah spent 20 to 25 hours speaking with authors to determine if they understood their own work, finding that many had no knowledge of the content in the papers they submitted. TMLR editors used these interviews as an unusual preliminary step before issuing desk rejections for a subset of the submissions.

The investigation highlights vulnerabilities in academic publishing where penalties for submitting AI-generated papers are seen as insufficient. Critics note that because some synthetic submissions are accepted while others are merely desk rejected, researchers face a high incentive to cheat to secure publications.