As LLMs grow bigger, they're more likely to give wrong answers than admit ignorance

A study by Spanish researchers reveals that as large language models (LLMs) like BLOOM, LLaMA, and GPT become more advanced, they are less likely to admit when they don't know an answer. While overall accuracy improved with newer versions, the LLMs were more prone to guessing rather than acknowledging limitations. The study also found that even easy questions sometimes yielded incorrect responses, highlighting ongoing reliability issues. Additionally, human volunteers struggled to identify incorrect answers given by the AI models.