Profile banner
Rafael Cosman

Rafael Cosman

Founder & CEO, AI researcher, and entrepreneur in blockchain and education

rafaelcosman
San Diego, California
Israel
Joined March 2026
Favicon imagelinkedin.com
Favicon imagex.com
Favicon imagerafaelcosman.com
Favicon imagelesswrong.com

Network

24.7K connections
🌱
Startup Founders
🤖
AI/ML Innovators
📊
Operations & Strategy
💰
Venture Capitalists
🚀
Protogon Research Team
🎓
Stanford Academics
🪙
Digital Asset Professionals

Summary

Rafael Cosman is deeply committed to the advancement and alignment of artificial general intelligence (AGI), serving as the Founder & CEO of Protogon Research, an independent AI research group dedicated to exploring novel approaches in this field. His active participation and insightful contributions on LessWrong, a prominent platform for rationality and AI safety discussions, further highlight his intellectual engagement with the ethical and technical challenges of advanced AI. nitter+5
With a strong entrepreneurial background, Rafael co-founded Archblock (formerly TrustToken), a significant player in the decentralized finance (DeFi) space. He held key leadership roles, including CEO and Head of Engineering, where he was instrumental in developing TrueUSD, a collateralized stablecoin, and TrueFi, a non-custodial lending protocol, demonstrating his expertise in blockchain technology and financial innovation. muraena+2
Rafael is a passionate advocate for tech education and community development, exemplified by his co-founding of StreetCode Academy and CodeCamp. These initiatives, particularly in East Palo Alto, aimed to empower underserved youth with coding skills and foster an entrepreneurial culture, successfully raising over $1 million for their programs and significantly impacting student lives. muraena+5
His early career showcases a foundation in machine learning and engineering, including an internship at Palantir Technologies where he contributed to a patented crime prediction system (U.S. Patent No. 9,129,219). He also held Machine Learning Engineer roles at Google and as a Fellow at Lightspeed Venture Partners, further cementing his technical acumen and contributions to the field. muraena+3
Rafael's academic pursuits at Stanford University in Computer Science culminated in his graduation in 2015. During his studies, he demonstrated a keen interest in theoretical computer science and social choice, co-authoring a paper on a 'Three-Dimensional Framework for Locating Voting Rules' presented at the Public Choice Society, and teaching courses on voting theory and computability. muraena+5

Work

Education

Projects

Writing

Prototype of Using GPT-3 to Generate Textbook-length Content

January 18, 2023

LessWrong post showcasing a prototype that uses GPT-3 to generate long-form content, with a focus on educational materials.

Favicon imagelesswrong.com

What is wrong with this approach to corrigibility?

July 12, 2022

LessWrong post discussing an AI alignment approach where an AI is given a reward equal to its expected discounted future reward upon shutdown, exploring the implications and potential pitfalls of such a system.

Favicon imagelesswrong.com

Crime risk forecasting

September 8, 2015

U.S. Patent (No. 9,129,219) detailing models to predict the time and location of future crimes using ensembles of neural networks, support vector machines, and logistic regressions, developed during his internship at Palantir Technologies.

Favicon imageiq.wiki

A Three-Dimensional Framework for Locating Voting Rules

March 1, 2014

Co-authored publication with Nicolaus Tideman and James Green-Armytage, generalizing seven different voting rules into a single continuous space and exploring trade-offs between voting rules.

Favicon imageiq.wiki