Dec 01, 2024 - Georg Zoeller (AILTI) The Curious Case of David Mayer, the man ChatGPT cannot name.
ChatGPT users have noticed a curious behavior: It refuses to talk about a 'David Mayer'. We have the explanation and point out less obvious implications. TL;DR: While not a juicy conspiracy, OpenAI cut some corners on regulatory compliance and the implications are less benign than they seem. In fact, they can be weaponized with unexpected consequences. ChatGPT is well known for not knowing when to stop talking - the underlying transformer architecture lends itself to hallucinations in situations when the model is asked to generate text beyond the context it was trained on. Even more curious, sometimes it starts giving a response, only to change its mind mid sentence, and terminating the conversation. So naturally, when the software stops and refuses to answer, users take notice. In this case, ChatGPT users found that mention of the name “David Mayer”, whenever included in a message, would consistently cause the model to terminate the conversation. It’s a conspiracy! Creating even more mystery, chatGPT rejection messages quickly move from unhelpful to ominously threatening when the user starts investigating the phenomenon. Continua qui: <https://centreforaileadership.org/resources/analysis_the_curious_case_of_one_david_mayer/>
