Bill Sun
Chinese name: Qingyun Sun
Stanford Math PhD · AI researcher · investor · founder
Research profile
Stanford Ph.D. in the mathematics of AI. Early transformer contributor and the first researcher at Google Brain to make the Transformer work on QA (2016). A decade across foundational deep learning, large-scale optimization, and production ML in finance.
Published at ICML, NeurIPS, AAAI, CVPR, and CoRL on optimization, training dynamics, generative modeling, and decision-making under uncertainty.
Research roots
Stanford Math PhD focused on the mathematical structure of AI and new model architectures.
Industry track
Google Brain, Millennium, Citadel, Point72 Cubist, then startup building.
Community
Founding member of AGI House and co-builder of AI+.
Multi-agents
Protocol and new market design: AI agents, capital markets, smart contracts, poker, trading.
About
From AI structure to financial intelligence.
Current agenda
Three problems I keep returning to.
01
Recursive self-improvement
Make AI research both verifiable and guided by research taste.
02
Continual learning
Design better agent harnesses and RL in real environments with continuous domain shift, with financial markets as a primary example.
03
Multimodal pretraining
Build a financial world model pretrained on time series, events, spreadsheet or tabular data, microstructure, and other continuous signals.
Research
Five themes that shape the work.
01
Model architecture
Early transformer work at Google Brain, including the first QA result beyond translation.
02
Representation learning
Sparse representations, high-dimensional structure, and unsupervised learning.
03
Stochastic optimization
Optimization under noise, scale, and long training horizons.
04
Decision-making under uncertainty
RL, stochastic control, portfolio optimization, and robust action under delayed feedback.
05
Multi-agent
Protocol design and market structure for agentic capital.
Timeline
Main arcs, reduced to essentials.
2023-now
Founder mode
Building AI startups for investing, trading research automation, and machine-native capital markets.
2020-2023
Millennium
Built an AI-powered forecast trading team at Millennium-WorldQuant, after earlier work at Citadel AI and Point72 Cubist.
2016
Google Brain
Worked on early transformer-era QA research during the formative phase of modern attention models.
2014-2019
Stanford Mathematics PhD
Worked on optimization, sparse structure, inverse problems, investment modeling, and stochastic methods with David Donoho and Stephen Boyd.
2015-now
Blockchain and decentralized systems
Long-term work on Bitcoin, stablecoins, smart contracts, and agent-native capital protocols.
Network
Research is individual. Ecosystems are not.
Bill Sun's Podcast
Hosted and guest conversations across AI, crypto, robotics, startups, and market structure.
AGI House
Founding member of a Bay Area AI builder network linking researchers, founders, and early operators.
AI+ Club
Early AI builder community co-organized with Tim Shi and other founders.
Links