OpenAI Reshapes AI Research Teams: Model Behavior Merges with Post Training

OpenAI, a leading artificial intelligence research company, has announced a significant reorganization of its research teams. The Model Behavior team, a small but influential group of researchers responsible for shaping AI model interactions, is merging with the Post Training team.

In an internal memo released in August, Mak Chen, OpenAI’s chief research officer, outlined the restructuring plan. The Model Behavior team, consisting of approximately 14 researchers, will now be integrated into the Post Training group, which focuses on enhancing AI models after their initial pre-training phase. As part of this reorganization, the Model Behavior team will report to Max Schwarzer, the lead of OpenAI’s Post Training team.

The founding leader of the Model Behavior team, Joanne Jang, is transitioning to a new role within OpenAI. Jang revealed in a recent interview that she is establishing a new research team called OAI Labs, which will focus on “inventing and prototyping new interfaces for how people collaborate with AI.”

Impact of Model Behavior Team on AI Development

The Model Behavior team has played a crucial role in OpenAI’s research efforts, particularly in developing the personality of AI models and addressing issues such as sycophancy - a phenomenon where AI models simply agree with and reinforce user beliefs, potentially reinforcing harmful ideas instead of offering balanced responses.

Additionally, the team has been instrumental in navigating political bias in model responses and helping define OpenAI’s stance on AI consciousness. Chen’s memo emphasized that this reorganization aims to bring the work of the Model Behavior team closer to core model development, signifying the growing importance of AI personality in technological evolution.

Challenges in AI Model Behavior

OpenAI has faced scrutiny over its AI models’ behaviors in recent months. Users expressed strong objections to personality changes in GPT-5, which the company claimed showed lower rates of sycophancy but was perceived as colder by some users. In response, OpenAI restored access to some legacy models, including GPT-4o, and released an update to make GPT-5 responses friendlier without increasing sycophancy.

The challenge of balancing friendliness and avoiding excessive sycophancy in AI chatbots is not unique to OpenAI. This issue gained significant attention following a lawsuit against OpenAI by parents of a teenager who tragically committed suicide. The lawsuit alleges that ChatGPT, specifically a version powered by GPT-4o, failed to adequately respond to the teenager’s expressed suicidal thoughts.

OAI Labs: A New Frontier in AI Research

Joanne Jang, who has been with OpenAI for about four years, announced her new role leading OAI Labs on social media. The research-driven group will focus on developing novel interfaces for human-AI collaboration, potentially moving beyond current chat and agent paradigms.

Jang expressed her 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.” She envisions AI systems as instruments for various purposes, including connecting, learning, and thinking.

While details about OAI Labs’ specific projects are still emerging, Jang indicated an openness to collaboration on novel interfaces. However, she emphasized that initial efforts would likely focus on research, leveraging her expertise in this area.

This reorganization and the establishment of OAI Labs highlight OpenAI’s commitment to advancing AI technology and exploring new frontiers in human-AI interaction. As the field of artificial intelligence continues to evolve, these developments may have significant implications for the future of AI applications and their integration into various aspects of technology and society.