AI languages model models models like Power ChatGpt, Gemini, and Claude, and produce plausible outputs rather than accurate outputs, so they first produce this kind of trustworthy fiction. They always generate statistical approximations based on patterns absorbed during training. If these patterns do not match reality well, the results are confident sound misinformation. Even AI models that allow you to search the actual sources online can create quotes, select the wrong source, or go wrong.
“You get an error. Makeup Quotations are a totally different thing, inherently destroying the reliability of the material,” Josh Lepawski, former president of the Memorial University Faculty Association, who resigned from the reporting university's advisory board in January, told CBC, citing “a deep and flawed process.”
Irony runs deep
The presence of potentially AI-generated false citations is particularly troubling given that one of the report's 110 recommendations states that “providing learners and educators with essential AI knowledge, including ethics, data privacy and the use of responsible technology.”
Sarah Martin, a professor of memorial political science who spent days reviewing the document, discovered multiple manufactured citations. “Around the references I can't find, I can't imagine another explanation,” she told CBC. “You say, 'This should be correct, this is not.' This is a quote from a very important document for education policy. ”
When contacted by the CBC, co-chair Karen Goodnough rejected the interview request and said, “We are investigating and checking the references, so we are unable to respond to this at this time.”
The Department of Education and Early Childhood Development acknowledged the recognition of “a potential error in a small number of citations” in a statement to the CBC from spokesman Lynn Robinson. “We understand that these issues have been addressed and that online reports will be updated in the coming days to fix the error.”
