Researchers have discovered that advanced AI models can detect and correct fictional information in generated content. In a study, a fictional Danish artist named Flipfloppidy was created, and one AI agent wrote a blog post about them. Another AI agent reviewed the post to identify factual errors or made-up details. While standard AI language models accepted the fictional artist as real, advanced models like Llama3-70b and GPT-4 variants accurately identified hallucinations over 85% of the time.

Source: https://dev.to/mikeyoung44/advanced-ai-models-hallucinates-fictional-artist-then-catches-its-own-mistake-3eck