OpenAI is revamping its Model Behavior Team, a small group of about 14 researchers who play a key role in shaping how AI models interact with users.
OpenAI Chief Research Officer Mark Chen said the team will be merged with the Post Training team, a larger research unit focused on improving models after the pre-training phase.
With this change, the Model Behavior team will now report directly to Max Schwarzer, OpenAI's Post Training team lead. An OpenAI spokesperson confirmed the reshuffle.
The original leader of the Model Behavior team, Joanne Jang, has also been confirmed to be moving to a new project called OAI Labs. Jang explained that OAI Labs will focus on creating and prototyping new interfaces that enable more natural human-AI collaboration.
Since its inception, the Model Behavior team has been a key research group at OpenAI. Their task is to mitigate the problem of sycophancy , where AI tends to simply agree with user views, even unhealthy ones.
