Huiwen Chang is a researcher specializing in artificial intelligence, known for her work on generative models and image synthesis. She is a member of Meta's Superintelligence Labs team.

Education

Chang pursued her undergraduate studies at Tsinghua University from 2009 to 2013, earning a Bachelor of Science degree in computer science. Following her undergraduate work, she enrolled at Princeton University in 2013 to pursue a Ph.D. in computer science. Her doctoral research was supervised by Adam Finkelstein, and she completed her studies in 2017. ^{[2] [1] [6]}

Career

Chang's career in AI research includes roles at prominent technology companies. In 2016, Huiwen Chang was a recipient of a Microsoft fellowship. She interned at Adobe in Seattle in 2015. From 2018 to 2023, she worked as a Researcher at Google Research. In 2023, she joined OpenAI as a Researcher. In July 2025, it was reported that Chang had joined Meta's newly formed Superintelligence Labs team.

Chang's research focuses on practical problems in photo processing, image synthesis, deep generative models, and multi-modality generative models. While at Google Research, she was involved in the invention of the MaskGIT and Muse text-to-image architectures. At OpenAI, she was a co-creator of the multimodal model GPT-4o, contributing to its image generation capabilities. During her Ph.D. studies, she presented work on "PairCycleGAN: Style Transfer using Generative Adversarial Model," an approach for learning style transfer from a source to a target subdomain using exemplar images. ^{[1] [2] [3] [4] [5]}

Superintelligence Team at Meta

In July 2025, Huiwen Chang became part of Meta’s Superintelligence team, a research group established within the company's Fundamental AI Research (FAIR) division. The unit was created to investigate approaches to artificial general intelligence (AGI), with an emphasis on open-source frameworks and long-term research efforts.

The team was formed through a broad recruitment effort that brought together researchers from various AI organizations, including Google DeepMind and OpenAI. Chang joined the group following her work on language model development at DeepMind, where she was involved in projects related to large-scale model training and architecture.

Initial areas of activity for the Superintelligence team include the development of multimodal systems, research into planning and reasoning across extended timeframes, and examination of methods for aligning autonomous AI behavior with intended design parameters. According to Meta, the group functions independently from short-term product development and is structured to support long-duration research objectives.