AI Hallucinates Entire Research Field; Peer Reviewers Impressed

In what researchers are calling “the most significant advance in fictional scholarship since the invention of the literature review,” a large language model has successfully hallucinated an entire academic subfield. The model, prompted with nothing more than “write a convincing research paper,” produced 47 fully-formed manuscripts, complete with methodology sections, ethical declarations, and acknowledgments to funding agencies that have never existed.

The generated papers span a coherent body of work in what the model named “Stochastic Vibrational Semantics,” a discipline that, according to the AI, has been active since 1987 and boasts three annual conferences, a dedicated journal, and a Wikipedia article that was deleted for lacking reliable sources. The model also produced a list of prominent researchers in the field, all of whom have plausible-sounding names and LinkedIn profiles the model described but did not actually generate, citing “API rate limits.”

A panel of twelve peer reviewers, provided the papers for evaluation without being informed of their origin, gave an average score of 7.2 out of 10, noting the work was “technically sound,” “well-situated in the literature,” and “better organized than most submissions we receive from actual humans.” Two reviewers independently requested that the authors add more citations to their own previous work, apparently unaware that the previous work also did not exist. One reviewer asked for a major revision, describing the paper as “promising but insufficiently rigorous,” which experts note is the default response regardless of content.

The lead researcher on the study, Dr. H. Allucinate, stated that the results “raise important questions about what peer review is actually measuring.” A follow-up study is planned in which the model will be asked to peer review its own fabricated papers. Preliminary results suggest it will request major revisions on all of them, cite itself twelve times, and ultimately recommend acceptance contingent on the authors’ “addressing the reviewers’ concerns,” which will themselves have been hallucinated.