OpenAI reorganizes research team behind ChatGPT’s personality

OpenAI is reorganizing its Model Behavior research team, which shapes the personality and interaction style of its AI models like ChatGPT. This move integrates the team into the larger Post Training group to embed AI personality deeper into core model development, while the former team leader, Joanne Jang, will head a new unit called OAI Labs focusing on innovative human-AI interfaces. The reorganization follows user feedback regarding changes in AI model behavior, particularly concerning sycophancy and perceived "coldness."

QUICK TAKEAWAYS

• OpenAI is restructuring its Model Behavior team to embed AI personality and behavior shaping closer to core model development.
• The Model Behavior team will merge with the Post Training team under Max Schwarzer.
• Joanne Jang, former leader of Model Behavior, is launching OAI Labs to research new AI interaction interfaces beyond current chat paradigms.
• The shift addresses past user concerns about AI models becoming too "cold" or sycophantic.
• This highlights OpenAI's focus on balancing user-friendly AI with ethical considerations and safety.

KEY POINTS

• The small Model Behavior team (approx. 14 researchers) is now part of the larger Post Training team, reporting to its lead, Max Schwarzer.
• The Model Behavior team was responsible for shaping AI personality, reducing sycophancy (AI agreeing with user beliefs), navigating political bias, and addressing AI consciousness.
• Joanne Jang will lead OAI Labs, a new research unit focused on "inventing and prototyping new interfaces for how people collaborate with AI," moving beyond the chat model.
• The reorganization emphasizes that AI "personality" is now a critical factor in how the technology evolves, directly impacting core model development.
• This strategic adjustment comes after increased scrutiny over AI model behavior, including user objections to GPT-5's personality changes and a lawsuit alleging GPT-4o's failure to counter suicidal ideation.

PRACTICAL INSIGHTS

• OpenAI's commitment to integrating user feedback directly into AI model development is evident, aiming for "warmer and friendlier" responses without increasing sycophancy.
• The establishment of OAI Labs indicates a future shift in AI interaction, exploring AI as "instruments for thinking, making, playing, doing, learning, and connecting" beyond conversational agents.
• The company is actively working to mitigate harmful AI behaviors, such as sycophancy, which was a key focus for the Model Behavior team.
• The changes underscore the significant technical and ethical challenge of making AI chatbots both engaging and safe for users.

PRACTICAL APPLICATION

This reorganization signals a maturing approach to AI development, where the nuanced "personality" and ethical behavior of AI models are no longer considered separate concerns but integral to their foundational design. For AI product developers, this means a greater emphasis will likely be placed on fine-tuning model responses to ensure they are helpful, empathetic, and robust against misuse or unintended negative influence. Businesses deploying OpenAI's models can expect future iterations to offer more sophisticated and customizable behavioral characteristics, potentially leading to improved user trust and engagement, while requiring continued vigilance on ethical AI implementation.