Profile banner
Hugh Zhang

Hugh Zhang

AI Research Scientist & Harvard PhD Candidate

San Francisco, CA
scale+1
Joined February 2026
Favicon imagehughbzhang.com
Favicon imagelinkedin.com
Favicon imagescholar.google.com
Favicon imagex.com
Favicon imagegithub.com

Network

2.4K connections
🚀
AI Startup Founders
🧠
Google DeepMind Researchers
🧑‍🎓
Harvard PhD Students
🎓
Harvard AI Leadership
💡
OpenAI Researchers
👨‍🏫
Stanford AI Faculty
📚
Stanford PhD Students

Summary

Hugh Zhang is a highly accomplished AI researcher specializing in large language models (LLMs), reinforcement learning, and game theory. He is currently a Research Scientist at Scale AI, focusing on evaluating and improving LLM reasoning, and is simultaneously a PhD Candidate in Computer Science at Harvard University, supported by prestigious fellowships. His work includes significant contributions to the CICERO project, the first AI agent to achieve human-level performance in Diplomacy, and numerous publications in leading AI conferences and journals. hughbzhang+4
He is a co-founder of The Gradient, a digital magazine dedicated to making AI knowledge accessible and fostering community discussion. Through The Gradient, he has published articles on AI policy, the vulnerabilities of speech recognition systems, and the importance of open-sourcing language models, demonstrating his commitment to broader AI education and ethical considerations. thegradient+4
Hugh's academic background includes a B.A. with Honors in Economics from Stanford University, where he received the Sobieski Prize for Creative Thinking, and he was Valedictorian of Valley Christian High School. His early professional experiences span roles at Google, Facebook AI Research, and Asana, showcasing a broad foundation in both theoretical and applied aspects of technology and AI. hughbzhang+2
A keen enthusiast of the game Go, Hugh cites witnessing AlphaGo's victory as a foundational moment for his interest in AI. This hobby highlights his engagement with strategic thinking and complex problem-solving, which are central to his research in AI and game theory. hughbzhang+2

Work

Education

Projects

Writing

AGI Is Not Multimodal

June 1, 2025

An article discussing perspectives on Artificial General Intelligence (AGI) and multimodality.

Favicon imagethegradient.pub

Humanity's last exam

January 1, 2025

An arXiv preprint discussing a potential benchmark for assessing advanced AI capabilities.

Favicon imagescholar.google.com

A careful examination of large language model performance on grade school arithmetic

January 1, 2024

A study introducing the GSM1k benchmark to assess LLM performance on mathematical reasoning, revealing potential overfitting to existing benchmarks and discussing data contamination.

Favicon imagescholar.google.com

Chain-of-Thought Reasoning is a Policy Improvement Operator

January 1, 2023

A paper exploring chain-of-thought reasoning as an operator for policy improvement in AI systems.

Favicon imagescholar.google.com

Human-level play in the game of Diplomacy by combining language models with strategic reasoning

January 1, 2022

A scientific paper detailing the development of an AI system capable of human-level performance in Diplomacy, combining language models with strategic reasoning.

Favicon imagescholar.google.com

Trading off diversity and quality in natural language generation

January 1, 2020

A paper presented at the EACL HumEval Workshop on balancing diversity and quality in the outputs of natural language generation models.

Favicon imagescholar.google.com

OpenAI: Please Open Source Your Language Model

February 1, 2019

An article advocating for OpenAI to open source their language models.

Favicon imagethegradient.pub

Unifying human and statistical evaluation for natural language generation

January 1, 2019

A paper exploring methods to combine human judgments and statistical metrics for evaluating natural language generation systems.

Favicon imagescholar.google.com

Speech recognition systems are now vulnerable to adversarial attacks

May 1, 2018

An article discussing the vulnerability of speech recognition systems to adversarial attacks.

Favicon imagethegradient.pub

Hobbies

A lifelong Go player, with his interest in AI sparked by AlphaGo's victory over Lee Sedol. hughbzhang

Enjoys writing and eating burgers at In-N-Out. thegradient

His Instagram indicates an interest in Go, and includes a post about his Stanford graduation. instagram