- Turf wars at OpenAI have seen leadership shake-ups on key teams, including post-training.
- The post-training team is currently led by Liam Fedus, a researcher who helped develop ChatGPT.
The recent chaotic power struggles and turf wars within OpenAI have resulted in leadership shake-ups on some of the company’s most important teams. Among them is the post-training team, which is responsible for preparing AI models for wide release and has had three leaders in the past six months.
Currently leading post-training is Liam Fedus, a machine learning researcher who was part of a group of researchers that developed ChatGPT.
“We were definitely surprised how well it was received,” Fedus said of the chatbot in the MIT Technology Review in 2023, “there have been so many prior attempts at a general-purpose chatbot that I knew the odds were stacked against us.”
Fedus was also one of seven OpenAI researchers who developed a group of advanced reasoning models known as Strawberry. These models, which can think through problems and complete tasks they haven’t encountered before, represent a leap forward for AI.
Fedus began his academic career studying physics at MIT where he worked on a directional dark matter detector. He took a brief detour onto Wall Street, working as an equity research associate at Fidelity, before completing an MS in physics at UC San Diego, where he collaborated with scientists at the large hadron collider in Switzerland. He also received a PhD in computer science from the University of Montreal, where he focused on machine learning.
He went on to become a research scientist at Google Brain, the outfit responsible for many of the technological advances that underpin AI models today. He joined OpenAI in 2022 and, in less than two years, has found himself in one of the company’s most important roles.
Post-training is the stage of the development process that involves taking trained machine learning models and improving their performance and efficiency, ensuring that they meet real-world standards.
While perhaps less exciting than conducting frontier research or envisioning futuristic doomsday scenarios, the post-training team is an essential chokepoint through which OpenAI products have to pass before they end up in the hands of the masses. That’s particularly important as OpenAI struggles through the growing pains of transforming itself from a scrappy research outfit to a global corporation and profitable business.
Fedus took over the team after the departure of Barret Zoph, who left the company last month alongside CTO Mira Murati and VP of Research Bob McGrew. High-profile safety researcher Miles Brundage announced his departure from OpenAI on Thursday.
It’s not known exactly why the three prominent leaders departed, but it follows months of internal disagreements about balancing safety and commercialization in the development of advanced AI.
The simmering conflict came to a head last November with the sudden ouster and rapid reinstatement of CEO Sam Altman. Zoph himself took over the team in August following the departure of co-founder John Schulman, who left to join rival Anthropic.
OpenAI didn’t respond to a request for comment.
