Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
TechCrunch has learned, Opina is reconstructing its model behavior team, which is a small but dominant team that shapes the company’s AI models how people communicate with people.
OpenAI’s Chief Research Officer Mark Chen, in August, said that the model behavior team-which is comprised of about 5 researchers-will join the postal training team responsible for the improvement of the company’s AI models after their preliminary pre-training.
As part of the change, the model behavior team will now report the Openai post training Lead Max Schwarzer. A spokesman for the OpenAI confirmed the changes in TechCrunch.
Joan Jang, the founding leader of the model behavior team, is also moving forward to launch a new project at the organization. In an interview with TechCunch, Jang says he is forming a new research team called OAI Labs, which “will be responsible for inventing new interfaces and prototyping for how people cooperate with AI.”
Model behavior team has become one of the main research groups in Openai, which is responsible for the formation of the personality of the company’s AI models and decreasing Psychosophyness – which happens when the AI models simply agree and strengthen the user’s beliefs, even unhealthy issues instead of supplying balanced reactions. The party also worked on navigating political bias in model reactions and helped the Opena determine its position AIThe
Chen said in the memo to the staff that it is time to bring the opening model behavior team closer to the main model development. By doing this, the company indicates that its AI’s “Personality” is now considered an important reason for how the technology develops.
In recent months, OpenC has faced an increasing investigation into the behavior of AI models. Users objected to the sightingly Personality change Made on GPT -5, which the company says that low rates of psychophyse, it seemed cool to some users. It managed to retrieve access to OpenAEK, such as some inheritance models such as GPT -4O Update New GPT -5 reactions to feel “warm and friendly” without increasing psychophyse.
TechCrunch event
San Francisco
|
October 27-29, 2025
OpenAI and all AI models developers need to walk on a fine line to make their AI chatbot talk to a fine line, but not psychopantic. In August, the parents of a 16-year-old boy OpenAI has been sued against Chatzept’s alleged role in their son’s suicide. The boy, Adam Rhine said some of his suicidal thoughts and the plans for the ChatzP (especially a version powered by GPT -4O), according to the court documents, in months until his death. The case has been alleged that GPT -4o failed to return to his suicide ideals.
The model behavior team has worked on each OpenAI model since GPT -4, GPT -4.5, and GPT -5. Before the unit started, the initial image of Dal-E2, before Jang, worked on projects like the generation to equipment.
Jang announced at X posts Last week that he was going to “start something new in Open”. The former chief of model behavior has been with the Openai for almost four years, and before that, Dal-E2, the primary image of the Openai, worked on projects like the generation to equip.
Jang informed TechCrunch that he would serve as the General Manager of the OAI Labs, which would report to Chain. However, this is the first day, and what the interfaces in this novel are still not clear, he said.
Jang said, “I am really excited to explore the signs that take our chat outfits, which are currently associated with companionship or even agents, where the autonomy is emphasized,” Jang said. “I was thinking [AI systems] Thinking, creating, playing, making, learning and connecting as instruments for connection ”
If the OAI Labs would like to know if this novel would cooperate with the head of the former Apple Design Johnny Iv – who Work now AI with a family on the AI hardware device with Openai – Jang said he was open to a lot of concepts. However, Jang said he would probably start with the research fields with which he was more familiar.
Update 4:35 pm: This story was updated to include a link to Jang’s post, which was published after this story was published. We have also made it clear the models that OpenAE model behavior team worked on.
