Will “AI Stings” Also Fail?
Strong incentives can overcome evidence of problems for years and years
We’re now entering the “stings” phase of debunking the quality and reliability of LLMs and AI in science and scientific publishing, with a variety of efforts, including some by yours truly.
It’s all reminiscent of the period of OA stings, with one of the best by John Bohannon when he was at Science.
- It also brings to mind the work of sleuths like Elisabeth Bik, whose work is being feted this week for having started a decade ago.
The recent sting about bixonimania joins others about dyslexia as a gastrointestinal disease, bovine colostrum claims in a major medical LLM, to unblinking citation of planted articles about functional medicine, pesticides, and fluoride and IQ.
Overall, these systems seem very easy to manipulate and prank.
More evidence of LLM’s unreliability continues to emerge, and more evidence of our bad practices due to OA publishing incentives also continues to emerge (thanks again, Cureus).
But do stings work? Or are they just a sideshow while the incentives driving the growth of a cancerous information system continue to pump?
They didn’t stop or slow down OA in a perceivable way.
Will they also fail to create doubts about continued applications of AI and LLMs in scientific journal article writing, reviewing, and promotion?