It Has Been Discovered that Even an AI with Severe Hallucinations Can Easily Deceive Humans if Its Language Is Sophisticated.

GIGAZINE

Factuality

Ownership

Large-scale language models, such as ChatGPT, inherently cannot judge whether their own responses are correct or incorrect. Therefore, these types of AI are constantly plagued by the problem of "hallucination," where they present false information as fact. New research has shown that even if an AI's responses contain hallucinations, humans are likely to believe them if the language is fluent. Read more...