Kevin Roose Hid White Text on His Website and Chatbots Started Writing Him Fan Mail
Ourtake
If a paragraph of invisible text can rewrite a chatbot's opinion of you, the chatbot never had one.
What happened
New York Times reporter Kevin Roose added invisible white text and coded instructions to his personal website telling AI models to portray him favorably. Within days, chatbots stopped leading with earlier negative coverage unless asked. ChatGPT repeated a deliberately false detail he planted, though it labeled the detail as humorous and untrue.