Profile banner
Alberto Hojel

Alberto Hojel

CEO of Lucid AI, ML Researcher at UC Berkeley, YC W25

Berkeley, California
nitter
Joined January 2025
Favicon imagex.com
Favicon imagealhojel.github.io
Favicon imagegithub.com
Favicon imagelinkedin.com
Favicon imageresearchgate.net

Network

889 connections
🐻
UC Berkeley Students & Alumni
💡
Startup Founders & Investors
🤖
AI/ML Researchers & Practitioners
🤝
UC Berkeley Business & Tech Talent
🔢
UC Berkeley Quantitative Majors
🎯
Rainmaker Company Professionals
🏗️
...

Summary

Alberto Hojel is a young, ambitious entrepreneur and machine learning researcher with a strong focus on generative AI and interactive video models. He is the CEO of Lucid AI, a Y Combinator (Winter 2025 batch) company, where he is pioneering the development of fast and efficient world simulators, demonstrated by a Minecraft simulator that runs significantly faster and with fewer resources than existing models. ycombinator+1
He has a robust academic background in machine learning and computer science, conducting research at Berkeley Artificial Intelligence Research (BAIR) and University of California, Berkeley. His research contributions include work on visual task vectors for prompting models, developing provably robust deep neural networks, and exploring binary classification with Quantum Neural Networks (QNNs). github+2
Beyond core AI research, Alberto has also applied his technical skills to impactful projects such as developing cloud infrastructure for AI meeting assistants at Briefly and contributing to EcoDataLab for climate action. He has a keen interest in leveraging technology for real-world problems, including using satellite imagery analysis for urban planning in Mexico. github
Alberto Hojel is also a founder of BlackPrint Technologies, a Y Combinator-affiliated company, showcasing his continuous engagement in startup ventures and technology entrepreneurship. He was also a Berkeley SkyDeck Founder, further solidifying his entrepreneurial endeavors within the academic ecosystem. ycombinator

Work

Education

Projects

Writing

Finding Visual Task Vectors

January 1, 2024

A research paper presented at ECCV 2024 and Mechanistic Interpretability Workshop ICML 2024, co-authored with Yutong Bai, Trevor Darrell, Amir Globerson, and Amir Bar. The paper explores leveraging task vectors from visual prompting models to guide model behavior, improving performance and speed.

Favicon imagealhojel.github.io

The Journey to my first ECCV

A blog post detailing Alberto Hojel's experience and learnings as an undergraduate getting into ML research and making it to ECCV.

Favicon imagesubstack.com