Marvin von Hagen
Co-founder of Poke and a visiting graduate researcher at MIT, working on AI alignment, foundation models, and the science of auditing the systems we are starting to trust.
About
I am an entrepreneur and AI researcher based in Cambridge, Massachusetts. I co-founded Poke, an AI personal assistant built by The Interaction Company of California, and I am a visiting graduate researcher at MIT focused on artificial intelligence and foundation models.
Before MIT, I studied Management & Computer Science at the Technical University of Munich, co-founded TUM Boring - a student tunnel-boring team - and worked at Tesla and other organizations. My early work exposing misalignment behaviors in large language models, including extracting Bing / Sydney's system prompt, was covered in Time and the Washington Post.
- AI alignment
- Foundation models
- AI auditing
- Artificial intelligence
Research
Black-box access is insufficient for rigorous AI audits
FAccT 2024/Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency/Cited 265 timesCannot or Should Not? Automatic Analysis of Refusal Composition in IFT / RLHF Datasets and Refusal Behavior of Black-Box LLMs
NeurIPS 2024/NeurIPS 2024 SFLLM Workshop/Cited 13 times
Selected work with co-authors including Max Tegmark and Dylan Hadfield-Menell at MIT. Full list on Google Scholar.
Experience
- Now
Co-founder
Poke - The Interaction Company of California
Building an AI personal assistant that works the way people do.
- Now
Visiting Graduate Researcher
MIT
Artificial intelligence and foundation models, with a focus on alignment and auditing.
- Prev
Management & Computer Science
Technical University of Munich
Co-founded TUM Boring, a student tunnel-boring team competing internationally.
- Prev
Earlier
Tesla & others
Hands-on stints across engineering and operations before research and founding.