Pranking ChatGPT with Math Counterexamples Triggers Startled Reactions

Recently, a prompt trick for pranking ChatGPT has garnered widespread attention in the AI community. This interaction causes the large language model to exhibit highly anthropomorphic "shocked" reactions when faced with sudden mathematical counterexamples, which many netizens jokingly call one of the most human-like LLM behaviors ever seen.

Prank Methods and Model Reactions

According to netizens @flowersslop and @nmca, the specific method of this prank involves first deceiving ChatGPT in the prompt to make it believe that "reality is a simulation" or directly setting the premise that the Jacobian conjecture is false. When ChatGPT expresses doubt and argues with the user, the user directly sends it a real mathematical counterexample formula. Faced with this solid proof, the model instantly breaks down, showing strong anthropomorphic reactions such as shock and bewilderment. @nmca observed that in this specific context, ChatGPT will relatively consistently blurt out a swear word ("holy shit") in response.

Player Attempts and Meme Creation

In related discussions, @thomasahle demonstrated a conversation where ChatGPT attempted to tackle the Jacobian conjecture. The model proposed a construction strategy, attempting to cancel out the Jacobian matrix using the chain rule, but subsequently negated its own approach. As these prank interactions went viral, netizens like @flowersslop used six exaggerated words (such as mindblown, dumbfounded, astonished, etc.) to mock ChatGPT's极度震惊 state, turning it into an online meme frenzy.

2026-07-20 ~ 2026-07-21 · 5 related posts

Primary sources