OpenAI withdraws new ChatGPT version – because it was too nice
San Francisco. A new version of the AI chatbot ChatGPT has been withdrawn for an unusual reason: it was too nice to users. The variant of the GPT-4o model, released just a few days ago, made overly flattering comments and was also described as submissive, the developer company OpenAI explained.
In recent days, examples of the ChatGPT version enthusiastically supporting even definitely bad ideas—like hugging a cactus—have been piling up online. Users were also initially showered with praise even for simple questions.
OpenAI justified its intervention by saying that such behavior could be unpleasant and disturbing for users. ChatGPT's "default personality" influences how people interact with it and whether they trust the program.
One trigger for the problem seems to be that the version was designed to appeal to users - and according to the data, it was always well received by them.
OpenAI plans to offer more ways to customize the software's behavior in the future—for example, by offering a choice of different personalities. Even today, you can steer the tone of voice somewhat with instructions.
ChatGPT is the chatbot that sparked the euphoria surrounding artificial intelligence more than two years ago. The software now boasts around 500 million weekly active users.
RND/dpa
rnd
