
The "Mentally Retarded" AI: How Training on Junk Data Creates Irreversibly Dumb LLMs
In a provocative experiment that has sparked debates across AI research circles, scientists from three prominent U.S. universities—Shuo Xing from Stanford University, Junyuan Hong from the University of California, Berkeley, and Yifan Wang from Carnegie Mellon University—deliberately sabotaged a large language model (LLM) by training it on low-quality "junk data." The result? An AI that exhibits profound intellectual deficits, akin to what the researchers describe as "mental retardation" in human terms.













