Thanks for checking out my low-budget homepage!
I am VP of Engineering at OpenAI, where I'm fortunate to work with incredible leads on research teams (generative modeling, Algorithms, Language, RL, etc.) as well as our Infrastructure and hardware Acceleration teams. I am also responsible for our Policy team, which considers issues in the development and governance of AI systems.
With our most senior research scientists, I help set a broader research vision for OpenAI and work with teams to choose aligned research directions. At OpenAI, I've done anything needed to ship results like GPT-2 and to get the team from ~30 to the ~120 or so we are today: starting new research areas, promoting and coaching managers, technical work for the Microsoft fundraise, starting the recruiting team, rolling out our first perf system, etc.
My main focus is research, but most of my career has revolved around near- and long-term impacts of AI on society. I’m interested in how the dynamics of AI development can shape outcomes for people, and I try to spend as much time as possible with ethics, safety, and policy researchers.
I have a certificate in CS from Worcester State from when I was 12, and a BS/BA in Applied Math and Political Science from Yale.
I enjoy building research organizations that pursue a combination of directed basic research focused on advancing methods, and large projects focused on solving open challenges in ML by any means possible.
In the near term, I'm interested in (1) generative models for all modalities; (2) learning a distribution of environments and tasks to meta-learn in; (3) the interactions between technology and policy in determining AI outcomes.
I was formerly Director of AI at Axon, where I ran Axon Research, a group of a dozen folks conducting ML research and policy work to advance transparency and accountability in policing via bodycam footage. We worked on video categorization to enable citizens to get bodycam footage faster, and for communities to hold officers accountable for poor behavior.
The ethics side of our work is incredibly important and I was responsible for the efforts at Axon to establish an AI Safety/Ethics Board and a blanket ban on facial recognition on body-worn cameras, later backed by the first report from the Ethics Board.
I started and was CEO of Dextro, a deep learning company in video categorization. We started in the media space, but were pulled into body-worn camera footage by the Chief Data Science office of the Obama White House in 2015. Body-worn cameras were rolled out as a way to provide accountability for officer misconduct, but there are so many of them producing so much footage every day that communities couldn't separate random footage from actual citizen interactions. In 2017, Axon acquired us and rebranded Dextro's product as Axon AI.
Real-time classification of live video
At Dextro, in 2015, we were the first to ship an API that could classify videos live in real-time. We were covered extensively in FastCompany and Wired, among others. Facebook and Twitter did not release solutions here until a year or two later.
Axon AI Ethics Board
I created an interdisciplinary board with representation from the EFF, Future of Humanity Institute, NYU Law, civil liberties, chiefs, and community to regulate the use of AI in law enforcement. The goal is to ensure that ML is used solely to improve transparency and accountability and to prohibit dual-use for surveillance or its use in a decision-making loop. The second report written by the Board calls for significant regulation in the use of ALPR.
Calculating ideological drift over time in Congress
In school, I was very interested in modeling the behavior of political agents and spent some time with Prof. Jacob Hacker to figure out how to quantitatively estimate the change in the ideological "center" of America over time. We used an embedding of Congressperson vote behavior and calculated homographies between sessions of Congress to derive some interesting results solely based off of votes on bills.