A Reddit user posted a altered version of the Ebbinghaus optical illusion to test ChatGPT's image analysis. The AI incorrectly asserted that the two orange circles were the same size, despite the modification that made one circle visibly larger. Even after a prolonged dialogue of about fifteen minutes, ChatGPT remained convinced of its answer and did not adjust its reasoning. The episode highlights concerns about the chatbot’s reliance on internet image matching, its resistance to corrective feedback, and broader questions about the reliability of AI tools for visual tasks.
Leia mais → 
