Jiahui Yu is an AI researcher and a member of Superintelligence Labs, Meta’s team focused on developing foundation models and advancing artificial superintelligence. He previously served as Head of Research, Perception at OpenAI. ^[1]

Education

Yu earned a Bachelor of Science degree in Computer Science from the University of Science and Technology of China. He later pursued and completed a PhD from the University of Illinois, Urbana-Champaign ^[1].

Career

Yu began his research career as an intern at Microsoft Research Asia from May 2015 to December 2015, where he worked on large-scale deep learning training systems. In early 2016, he joined Megvii (Face++) as a research intern, focusing on object detection and contributing to the development of UnitBox. He then interned at Adobe Research from May to December 2017, working on generative adversarial networks and image inpainting, including DeepFill v1 and v2. Between June 2016 and May 2018, he also served as a research consultant at Jump Trading LLC, where he developed price prediction models and trading strategies utilizing high-frequency financial data.

In 2018, Yu continued his research trajectory with Snap Research from January to August, developing efficient deep learning models, including Slimmable Networks. He later joined Baidu Research as an intern from August to December 2018, contributing to projects in AutoML and efficient neural networks like AutoDL 2.0 and EasyDL. In early 2019, he interned at Nvidia Research from January to May, focusing on large-scale generative models for visual content generation. This was followed by a research internship at Google Brain from May to September 2019, where he worked on AutoML and neural network efficiency, including BigNAS.

Yu joined Google in February 2020 as a research scientist, contributing to the development of streaming automatic speech recognition and next-generation multimodal embeddings used across Alphabet products. He later advanced to senior research scientist from May 2021 to October 2022, focusing on multimodal understanding and generation. From November 2022 to September 2023, he served as a staff research scientist at Google DeepMind, where he co-led the Gemini Multimodal project and contributed to the development of the PaLM-2 architecture.

In October 2023, Yu joined OpenAI as Head of Research for the Perception team, where he currently leads efforts in areas related to perception-based AI research. In July 2025, he joined the Meta Superintelligence Team, Meta’s unified division focused on developing foundation models, advancing AI research, and building toward artificial superintelligence. ^{[2] [6]}

Superintelligence Labs (MSL)

Superintelligence Labs (MSL) is a division within Meta, launched in June 2024, to unify and accelerate the company’s artificial intelligence initiatives, particularly in pursuit of artificial general intelligence (AGI). Led by Alexandr Wang and Nat Friedman, MSL brings together teams working on foundation models, applied AI products, and core research from FAIR. The unit was formed alongside a major talent acquisition campaign, hiring researchers from OpenAI, Anthropic, and DeepMind, and follows Meta’s $14.3 billion investment in Scale AI. MSL oversees the development of Meta’s Llama model series and next-generation AI systems, with a focus on long-term advancements and integration across Meta’s consumer platforms. ^{[2] [3]}

Interviews

In a past episode of "Meet a Google Researcher," host Drew Calcagno interviewed research scientists Yonghui Wu and Yu about Google's text-to-image model, Parti. They discussed how Parti, an autoregressive model, generates high-quality images from text prompts by treating the image generation process like machine translation. The researchers highlighted the model's ability to produce intricate visuals based on detailed descriptions, while also contrasting it with Google's other model, Imagen, which uses a diffusion approach. They shared their experiences of developing Parti, including moments of breakthrough during debugging sessions and the excitement of witnessing significant improvements as they scaled the model. Both researchers expressed their anticipation for future advancements, including the integration of different modalities like video and music, while emphasizing the importance of responsible AI practices in making these technologies accessible. ^[4]