Shuchao Bi

Also known as 毕树超

Shuchao Bi is an AI researcher and lab leader whose work links recommendation systems, video discovery, and multimodal model training. He studied mathematics at Zhejiang University and graduate statistics and mathematics at UC Berkeley before building advertising recommendations at Google and leading discovery for YouTube Shorts. At OpenAI, he led multimodal post-training and co-created GPT-4o voice mode and o4-mini; since 2025, he is a researcher at Meta Superintelligence Labs. His papers explore new user interests with language models, knowledge transfer into specialized models, and token-level collaboration among language models.

Now at: Meta Superintelligence Labs
Works: 12
Citations: 75
h-index: 5

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

By 2013

Mathematics at Zhejiang and Berkeley

Before joining Google in 2013, he studies mathematics at Zhejiang University, earns a master's degree in statistics, and pursues a doctorate in mathematics at UC Berkeley.
2013–2019

Google advertising recommendations

From 2013 to 2019, he serves as a technical leader at Google and builds a multistage deep-learning recommendation system for its advertising business, which significantly increases Google's advertising revenue, in the order of billions of dollars.
2019–2024

YouTube Shorts discovery

From 2019 to 2024, he heads exploration for YouTube Shorts, co-creates and leads its video recommendation and discovery system, and builds a large machine-learning team across recommendations, scoring, interaction discovery, trust, and security.
2024

LLMs for user interest exploration

In 2024, he coauthors LLMs for User Interest Exploration in Large-scale Recommendation Systems, a hybrid framework that combines large language models with classic recommendation models to recommend novel user interests.
2024

Wisdom of committee

In 2024, he coauthors Wisdom of Committee: Diverse Distillation from Large Foundation Models and Domain Experts, which finds that adding domain-specific experts to foundation-model teachers improves knowledge transfer to compact domain models.
2024–2025

OpenAI multimodal post-training

After joining OpenAI in 2024, he leads the multimodal post-training organization and co-creates GPT-4o voice mode and o4-mini.
Since 2025

Research at Meta Superintelligence Labs

He joins Meta Superintelligence Labs in 2025 and is a researcher there.
January 2026

FusionRoute token-level collaboration

In January 2026, he coauthors FusionRoute, a token-level framework that selects an expert model at each decoding step and adjusts its next-token distribution.