Jack Kunde
I am building technology that seems infeasible in the short term, but, when developed, will have a massive positive impact on humanity.
My primary interest is robotics. By developing general purpose robotics models deployable in any industry, we have the chance to reduce the cost of goods for everyday Americans, bring essential industries back to the US, and create jobs and entire fields of work that never existed before.
To get there, I am particularly interested in the new techniques and data sources for robot learning, and in working to understand manufacturing and other valuable use-cases for robotics.
At Ludo Robotics, I am learning about building foundation models for robotics in the best way — by actually doing it. Previously, I worked on machine learning, software, and, occasionally, hardware at an agriculture startup called Ohalo. I have a degree in math and computer science from the University of Wisconsin, where I worked with Professor Kangwook Lee researching language models.
We are on the cusp of something great, and there has never been a more exciting time to follow my curiosity as I learn as much as possible, as fast as possible.
If any of this is interesting to you I'm always available to talk. Send me a message by email jkunde at wisc dot edu or on Twitter @kunde_jackson.
Publications
Synthetic data to improve PRMs and scale test-time compute on new domains. ICML ’25 Spotlight.
A queuing-theory approach to increasing throughput of LLM systems.
Opinion piece on computer-assisted proofs, from early computers to language models.
Projects
Demoed to members of Congress and their staffers a multi-agent system that places realistic AI phone calls, showing how false emergency reports could be automated at scale.
RLHF using a custom transformer and a policy gradient methods.
Decoder-only language model in PyTorch, trained on Shakespeare.