...“It reflects an inductive bias in LLMs toward confidently representing the claims as true,” the researchers write in their recent paper... The illusory truth
Researchers have identified an "illusory truth effect" in large language models, where the models tend to confidently present false statements as true. This phenomenon mirrors a similar cognitive bias observed in humans. The study suggests that LLMs possess an inductive bias that favors asserting claims with certainty, even when those claims have been explicitly flagged as false. AI
IMPACT Highlights a potential vulnerability in LLMs related to truthfulness and confidence, impacting their reliability in information dissemination.