Shengjia Zhao

Shengjia Zhao took a PhD in computer science at Stanford in Stefano Ermon's group, working on generative models and probabilistic machine learning. An early paper, InfoVAE (2017), proposed a way to stop variational autoencoders from ignoring their latent variables, a common failure of the method.

After Stanford, Zhao joined OpenAI and worked on the models that made the company's name: Mark Zuckerberg's list of credits called Zhao a "co-creator of ChatGPT, GPT-4, all mini models, 4.1 and o3" who had led OpenAI's synthetic data work, and TechCrunch reported that Zhao had also contributed to o1, the first OpenAI model trained to reason at length before answering.

Meta hired Zhao in June 2025 as one of the first recruits to Meta Superintelligence Labs, and on 25 July Zuckerberg named Zhao its chief scientist, writing that Zhao had "co-founded the new lab and has been our lead scientist from day one." The appointment has two accounts. The Financial Times reported in August 2025 that within days of arriving Zhao had threatened to go back to OpenAI, even signing the paperwork, and was then given the chief scientist title. Meta said Zhao had been slated for the role from the outset.

As chief scientist Zhao sets MSL's research agenda under Alexandr Wang. When the lab's first model, Muse Spark, was released in April 2026, Zhao posted "Meta is back!" and pointed to its token efficiency on third-party evaluations. In September 2026 Zhao announced the Muse personal agent, stressing a security design in which each agent runs in its own virtual machine and a separate monitor checks every action.