ChatGPT's Voldemort: Chatbot can't write David Mayer

If you ask ChatGPT who David Mayer is, the AI chatbot says that it is unable to answer. Nor can it write the name. Even with various attempts to trick it and convince it to do so, ChatGPT refuses to do the job. Attempts by heise online to elicit a "Mayer" from him have failed. ChatGPT obviously has a Voldemort – that is the villain in Harry Potter whose name nobody pronounces.

The peculiarity was noticed by a Reddit user. Numerous other people then tested whether they would get the same answer. And indeed: ChatGPT has the same difficulty with David Mayer for everyone. Now the question arises as to why this is the case.

Conspiracy, legal dispute or coincidence?

If you google David Mayer, David Mayer de Rothschild comes up first. The search engine has saved this Mayer in the Knowledge Graph as an adventurer and ecologist, and as an expert on raising awareness of climate change. So far, ChatGPT has not attracted attention as a climate change denier – and also not necessarily with anti-Semitic conspiracy narratives about the Rothschilds, a banking dynasty from which this David Mayer de Rothschild also originates. ChatGPT has no further problems with the name Rothschild itself.

David Mayer is not a particularly rare name. Incidentally, ChatGPT can spell David and Mayer individually very well and reliably. So why does the chatbot even warn some people that further questions about the name could be a violation of the usage guidelines? One user is said to have been told by ChatGPT that the name closely matches a "sensitive or marked entity" and that personal rights or the rights of public figures or brands must not be violated.

This has further fueled the rumors surrounding the Rothschild David. However, there is also another David Mayer that is being traded as a reason: A wanted criminal is said to have once posed as a David Mayer in the USA, causing a man with the same name who lived in the UK to be wanted – and blacklisted. Because of this mistake, the real and innocent David Mayer could not travel to his home country, the USA, and could not receive letters from there, it is said.

However, some believe that there could also be a David Mayer who is in dispute with OpenAI over copyright or other personal rights, which is why the company initially blocked the name. The GDPR stipulates that false information about individuals must be deleted if this is requested. There is the right to be forgotten, according to which you can request to be removed from Google search results. It is still unclear how this will be handled for AI models in the future. No information can be deleted after training; the models can only be prohibited from giving certain answers by means of fine-tuning.

On X, a user has found other names that apparently cause similar difficulties for ChatGPT. Among them is an Italian lawyer who says he has invoked the right to be forgotten and demanded that OpenAI stop ChatGPT from talking about him.

What all these theories have in common is that OpenAI actively ensured this ban. But this is not necessarily the case. Part of how AI models work is still a kind of black box. They are trained with gigantic amounts of data and link the content in a structure: artificial neural networks are created. However, this can also result in connections that are incorrect. This can lead to hallucinations, for example. In addition, training data may have been deliberately poisoned. This means that false information can be fed to the AI. This can also lead to ChatGPT thinking that David Mayer is an unpronounceable name.

(emw)