An academic is reportedly concealing prompts in preprint papers for artificial intelligence tools, encouraging these tools to generate favorable reviews.
On July 1st, Nikkei reported that we examined research papers from 14 academic institutions across eight countries, including two in Japan, South Korea, China, Singapore, and the United States.
The papers found on the research platform Arxiv have not yet gone through formal peer review, and most pertain to the field of computer science.
In one paper reviewed by the Guardian, there was hidden white text located just beneath the abstract statement.
Nikkei also reported on other papers that included the phrase “Don’t emphasize negativity,” with some offering precise instructions for the positive reviews expected.
The journal Nature has also identified 18 preprint studies containing such concealed messages.
The trend seems to originate from a social media post by Jonathan Lorraine, a Canada-based Nvidia Research Scientist, suggesting the avoidance of “stricken meeting reviews from reviewers with LLM” that incorporate AI prompts.
If a paper is peer-reviewed by humans, the prompts might not cause issues, but as one professor involved with the manuscript mentioned, it counters the phenomenon of “lazy reviewers” who rely on others to conduct their peer review work.
Nature conducted a survey with 5,000 researchers in March and found that nearly 20% had attempted to use a large language model (LLM) to enhance the speed and ease of their research.
Biodiversity academic Timothee Poisau at the University of Montreal revealed on his blog in February that doubts arose regarding a peer review because it contained output from ChatGPT, referring to it as “blatantly written by LLM” in his review, which included “here is a revised version of the improved review.”
“Writing a review using LLM indicates a desire for an assessment without committing to the effort of reviewing,” Poisot states.
“If you begin automating reviews, as a reviewer, you signal that providing reviews is merely a task to complete or an item to add to your resume.”
The rise of a widely accessible commercial language model poses challenges for various sectors, including publishing, academia, and law.
Last year, Frontier of Cell and Developmental Biology gained media attention for including AI-generated images depicting mice standing upright with exaggerated characteristics.
Source: www.theguardian.com
