Just Now, Four Chinese Researchers from OpenAI Are Poached by Meta in Heavy Recruitment}
Meta has recruited four top researchers from OpenAI, including Chinese scholars, shortly after its recent large-scale hiring spree, intensifying competition in AI development.
Once again, Meta has 'poached' members from OpenAI. The last time Meta took over OpenAI’s Zurich office was just a few days ago.
After launching the Llama 4 AI model in April, Meta initiated a large-scale recruitment wave. The performance of Llama 4 did not meet CEO Mark Zuckerberg’s expectations, and Meta faced criticism for its version used in popular benchmarks.
Meanwhile, a war of words erupted between Meta and OpenAI. OpenAI CEO Sam Altman claimed Meta was offering "$100 million signing bonuses" to attract talent, but added that "our top talents have not been poached so far."
In response, Meta CTO Andrew Bosworth told employees that while some executives received similar offers, the actual terms were much more complex than a one-time bonus. In other words, it’s not just a quick cash deal.
The last three colleagues who were poached had participated in key research like ViT. The new recruits also contributed to many important OpenAI projects.
- Jiahui Yu: Led the development of o3, o4-mini, and GPT-4.1
- Hongyu Ren: Creator of o3-mini and o1-mini, and a core contributor to GPT-4o and GPT-4
- Shuchao Bi: Head of multimodal post-training organization at OpenAI
- Shengjia Zhao: Key contributor to GPT-4 and o1
These researchers are the backbone of OpenAI’s models from GPT-4 to GPT-4o, as well as lightweight models like o1-mini and o3-mini. It remains to be seen whether this will cause a talent gap at OpenAI and impact GPT-5 development. After absorbing these talents, Meta’s weakest link in large model tech—fine-tuning and multimodal alignment—may see a leap forward. Let’s keep an eye on it.
Netizens’ fun guesses about Llama 5
Next, let’s briefly review the backgrounds of these researchers:
Shengjia Zhao
According to LinkedIn, Shengjia Zhao joined OpenAI in June 2022. He graduated from Tsinghua University with a bachelor’s degree and earned his PhD in computer science from Stanford University, winning the ICLR 2022 Outstanding Paper Award.
After joining OpenAI, Shengjia Zhao participated in key large model training, including GPT-4, GPT-4o, and o1.
Jiahui Yu (余家辉)
Jiahui Yu joined OpenAI in October 2023 and is now head of the Perception team. Previously, he led multimodal projects at Google DeepMind Gemini.
He graduated from the University of Science and Technology of China’s Youth Class in computer science and earned his PhD from the University of Illinois at Urbana-Champaign under Professor Thomas Huang. His research covers deep learning and high-performance computing.
At OpenAI, he contributed to projects like "Thinking with Images," o3, o4-mini, GPT-4.1, GPT-4o, and image generation.
Shuchao Bi
Shuchao Bi joined OpenAI in May 2024 and is now head of Post-training Multimodal research. Previously, he was a tech lead at Google and director of engineering at YouTube.
He graduated from Zhejiang University and obtained his PhD from UC Berkeley.
His core research includes pretraining paradigms, multimodal reasoning, high-order reinforcement learning, multimodal scoring and evaluation systems, agent integration, multimodal-multilingual cognition, embodied intelligence, and multimodal distillation.
Hongyu Ren (任泓宇)
Hongyu Ren is a research scientist at OpenAI, joining in July 2023 after working at Apple and Google. He holds a PhD in computer science from Stanford and a bachelor’s from Peking University.
He contributed to the creation of o3-mini, o1-mini, and was a key contributor to GPT-4o and GPT-4. He also led a post-training multimodal team.