Revealed that even AI with terrible hallucinations can easily deceive people if the wording is sophisticated

Research✍ TextPublished: Source: Gigazine

#OpenAI

Key point

Large language models like ChatGPT cannot judge the correctness of their own answers, leading to the problem of "hallucination." New research found that even if hallucinations exist, humans are highly likely to believe them if the wording is fluent.

Large language models represented by ChatGPT and others, as a fundamental mechanism, cannot judge for themselves whether their own answers are correct or incorrect. Therefore, this type of AI is constantly haunted by the problem of "hallucination," where it speaks non-facts as if they were facts.

New research has found that even if there are hallucinations in an AI's answer, if the wording is fluent, there is a high possibility that humans will believe it.

This item is a summary. The full, checkable version is at the original source:

Read the original at Gigazine ↗