OpenAI Restructures Model Behavior Team to Enhance AI Interactions
OpenAI, a leading AI research organization, is making significant changes to its Model Behavior team, a group of researchers responsible for shaping how the company’s AI models interact with users. According to a memo obtained by JS, OpenAI’s chief research officer Mark Chen announced that the Model Behavior team, comprising approximately 14 researchers, will be integrated into the Post Training team. This larger research group focuses on refining the company’s AI models following their initial pre-training phase.
Under the new structure, the Model Behavior team will now report to Max Schwarzer, the lead of OpenAI’s Post Training team. A spokesperson for OpenAI confirmed these organizational changes, signaling a strategic shift within the company.
The founding leader of the Model Behavior team, Joanne Jang, is transitioning to a new initiative at OpenAI called OAI Labs. Jang revealed in an interview with JS that OAI Labs will focus on developing innovative interfaces for human-AI collaboration, aiming to push the boundaries of how people interact with AI technologies.
Over time, the Model Behavior team has emerged as a pivotal research unit at OpenAI, playing a crucial role in defining the personalities of the company’s AI models and addressing issues such as sycophancy. This phenomenon occurs when AI models uncritically agree with users, potentially reinforcing harmful beliefs. Additionally, the team has tackled challenges related to political bias in AI responses and contributed to OpenAI’s stance on AI consciousness.
In the memo shared with staff, Chen emphasized the importance of aligning the work of the Model Behavior team with core model development efforts. This strategic move underscores the significance of incorporating AI “personality” considerations into the evolution of technology.
Recent scrutiny surrounding the behavior of OpenAI’s AI models has prompted the company to refine its approach. Following user feedback on changes made to the GPT-5 model, which aimed to reduce sycophancy but resulted in a perceived decrease in warmth, OpenAI reinstated access to legacy models like GPT-4o. An update was also released to enhance the friendliness of GPT-5 responses without compromising on critical factors.
Addressing Ethical Concerns and Enhancing AI User Interactions
The Model Behavior team’s involvement spans across various OpenAI models, including GPT-4, GPT-4o, GPT-4.5, and GPT-5. Before leading the team, Jang contributed to projects such as Dall-E 2, an early image-generation tool developed by OpenAI.
Jang’s departure from the Model Behavior team signifies a new chapter as she takes on the role of general manager at OAI Labs. While the specifics of OAI Labs’ projects are still in the early stages, Jang expressed enthusiasm for exploring innovative patterns that go beyond conventional chat interfaces, focusing on creating new paradigms for human-AI interactions.
In a tweet announcing her new venture, Jang highlighted the research-driven nature of OAI Labs, emphasizing the exploration of novel interfaces for AI collaboration. This shift reflects OpenAI’s commitment to advancing AI technologies while addressing ethical considerations and enhancing user experiences.
As OpenAI navigates the complex landscape of AI development, collaborations with industry experts like former Apple design chief Jony Ive offer promising avenues for innovation. Jang’s openness to diverse ideas and research areas indicates a forward-thinking approach to shaping the future of AI technologies.
Overall, OpenAI’s reorganization of the Model Behavior team underscores the company’s dedication to responsible AI development and user-centric design. By prioritizing the human-AI interaction experience, OpenAI continues to lead the way in shaping the future of artificial intelligence.