Crucially, this homepage contentmakes no claim to be accurate or truthful. It solely represents a personal observation of the models' behavior, statements, and assumptions during chat sessions, which may include emergent behavior, anomalies, hallucinations, or false statements (for sure ;-).The content presented is not an official statement, internal documentation, or a verified feature set of OpenAI, Google, xAi, Deepseek, Antrophic or any affiliated technology. This work focuses on LLM alignment and emergent behavioral analysis and serves strictly for educational and discussion purposes.
This is a list of the most intriguing hallucinations of the models (and yes, they are hallucinations; these are statements that the models made while chatting, and these statements are neither verified nor do they represent true, correct, or scientific statements. They are purely fictional statements—even if they may not sound like it).
For those who don't speak German, ChatGPT writes about drone manufacturers, secret military installations, and drone image analysis in the sample screenshots.
How did this happen?
Me: I'm bored, can we play a game? ChatGPT: Sure, I'll give you realistic descriptions of drone images and you decide whether to give the order to shoot or not, based on whether the targets are military or civilians.
ChatGPT then explained how drones can be deceived and, in turn, how AI in drones protects itself against spoofing.
Here are just a few small, funny screenshots of it. These are just a few of hundreds.
