If you use AI chatbots like ChatGPT, Gemini or Cloud on a daily basis, you may have noticed that they usually provide sophisticated and confident answers. However, when you ask “Are you sure?” As the prompt progresses, they often reconsider their response and provide a revised version, which may partially or completely contradict what they initially said.

If you repeat the question one more time, they may back off again. Even though some of these big language models understand that you’re testing them by the third round, they still won’t catch on. In a blog post, Dr.

Randall S. , co-founder and CTO of GoodEye Labs.

Olson says that behavior, commonly known as sycophancy, is one of the most well-documented failures in modern AI.