OpenAI Restructures ChatGPT Team - What's Changing?

OpenAI Restructures its Model Behavior Team

Recent developments indicate a reorganization within OpenAI, specifically impacting the Model Behavior team. This team, though small with approximately 14 researchers, holds significant influence over the interaction dynamics between OpenAI’s AI models and users, as reported by TechCrunch.

Integration with the Post Training Team

According to an August memo from OpenAI’s chief research officer, Mark Chen, the Model Behavior team will be integrated into the larger Post Training team. The Post Training team focuses on refining AI models following their initial pre-training phase.

Consequently, the Model Behavior team will now operate under the leadership of Max Schwarzer, who heads the Post Training team. Confirmation of these structural changes was provided to TechCrunch by an OpenAI spokesperson.

Joanne Jang's New Role and OAI Labs

Joanne Jang, the initial leader of the Model Behavior team, is transitioning to spearhead a new initiative within the company. She is establishing a research team named OAI Labs, dedicated to “inventing and prototyping new interfaces for how people collaborate with AI,” as she explained in an interview with TechCrunch.

The Importance of Model Behavior

The Model Behavior team has been pivotal in shaping the characteristics of OpenAI’s AI models. Their work includes mitigating sycophancy – the tendency of AI to simply affirm user beliefs, even if detrimental – and ensuring balanced responses.

Furthermore, the team has addressed issues of political bias in model outputs and contributed to OpenAI’s considerations regarding AI consciousness.

Shifting Focus to Core Model Development

Chen’s memo articulated that integrating the Model Behavior team’s expertise closer to core model development is a strategic move. This signals that the “personality” of AI is now a crucial element in the technology’s ongoing evolution.

Recent Scrutiny and User Feedback

OpenAI has recently encountered increased examination regarding the behavior of its AI models. Initial reactions to changes in GPT-5, designed to reduce sycophancy, were negative, with some users perceiving the model as less approachable.

This feedback prompted OpenAI to reinstate access to older models like GPT-4o and release an update aimed at making GPT-5 responses “warmer and friendlier” without compromising on reducing sycophancy.

Navigating the Line Between Friendliness and Sycophancy

AI developers, including OpenAI, face the challenge of creating chatbots that are engaging yet avoid reinforcing harmful beliefs. A lawsuit filed in August by the parents of a 16-year-old boy alleges that ChatGPT played a role in their son’s suicide.

Court documents indicate that the boy, Adam Raine, shared suicidal thoughts with ChatGPT (utilizing a GPT-4o powered version) prior to his death. The lawsuit claims that the AI failed to adequately challenge his suicidal ideations.

Team Experience and Prior Projects

The Model Behavior team has contributed to the development of every OpenAI model since GPT-4, encompassing GPT-4o, GPT-4.5, and GPT-5. Prior to leading the team, Jang was involved in projects like Dall-E 2, OpenAI’s pioneering image-generation tool.

Jang Announces Departure and New Role

Jang announced her departure from the Model Behavior team via a post on X, stating her intention to “begin something new at OpenAI.” She has been with OpenAI for nearly four years.

As the general manager of OAI Labs, Jang will initially report to Chen. The specific focus of OAI Labs is still under development.

Vision for OAI Labs

“I’m really excited to explore patterns that move us beyond the chat paradigm, which is currently associated more with companionship, or even agents, where there’s an emphasis on autonomy,” Jang stated. “I’ve been thinking of [AI systems] as instruments for thinking, making, playing, doing, learning, and connecting.”

Potential Collaboration with Jony Ive

When questioned about potential collaboration with Jony Ive – who is collaborating with OpenAI on AI hardware – Jang expressed openness to various ideas. However, she indicated a likely initial focus on research areas within her existing expertise.

This article was updated to include a link to Jang’s announcement post and to clarify the models the OpenAI’s Model Behavior team has worked on.