OPENAI
ChatGPT’s vibe check
OpenAI is merging its Model Behaviour team, the small but influential group shaping how its AI models interact with people, into the larger Post Training group.
In a memo to staff, Mark Chen, OpenAI’s chief research officer, said the move will bring personality design closer to core model development.
The team of 14 will now report to Max Schwarzer, who leads Post Training.
Joanne Jang, the Model Behaviour team’s founding lead, is stepping away to launch OAI Labs, a new research unit exploring “novel interfaces” for human–AI collaboration.
She’ll stay at OpenAI and report to Chen as the lab gets underway.
The Model Behaviour team has played a key role in shaping OpenAI’s models, tackling issues like sycophancy (when AI agrees too much), political bias, and personality design across GPT-4, GPT-4o, GPT-4.5, and GPT-5.
This change comes at a time of heightened scrutiny.
What this means
The Model Behaviour team is joining OpenAI’s Post Training group to bring personality design closer to model development.
Joanne Jang is leading OAI Labs, focused on rethinking how humans interact with AI.
OpenAI is working to balance warmth with accuracy as it shapes its future models.
So how friendly is too friendly?
Users recently complained that GPT-5 felt “colder” despite being less sycophantic, prompting OpenAI to restore older models and tweak responses to make them “warmer and friendlier.”
Jang says OAI Labs will look beyond the standard chat format, describing AI systems as “instruments for thinking, making, playing, doing, learning, and connecting.”
Less people-pleasing, more personality tweaking.