Career
I cofounded the
AI Futures Project and work as a researcher there on AGI forecasting and governance.
I've worked on the
AI 2027 scenario and the
AI Futures Model, which predicts AI timelines and takeoff speeds.
I advise
Sage, an organization that is as of 2026, primarily focused on the
AI Village.
I led the creation of the
2025 and
2026 AI forecasting surveys.
I previously worked as a software engineer at
Ought on the AI research assistant
Elicit.
Selected work
Some work that I've (co-)authored that I am happiest about:
Improving outcomes from superhuman AGIs
I've decided to work on improving outcomes from superhuman AGIs
because I think it's both the most important problem we face, and furthermore I find it interesting.
If you'd like to help, I've written a post with advice:
What you can do about AI 2027.
I also consider myself part of the
effective altruism
movement, which aims to use reason and evidence to improve the world as
much as possible (this doesn't mean I endorse every action prominent effective altruists or effective altruism orginazations take!). I've taken the
Giving What We Can Pledge
to donate at least 10% of my lifetime income to whatever I think is the most effective use of my money,
and ideas I learned about due to effective altruism
have had a large impact on my career decisions.
Forecasting
I'm interested in judgmental forecasting as a route to improved decision
making. I recommend
Superforecasting
for an introduction, but I think it overclaims in some ways (e.g.
regarding
generalists vs. expertise). I co-lead the
Samotsvety Forecasting team. My
forecasting track record is described
here. Some takes on forecasting/epistemics interventions as of March 2024
here.
AI robustness research
During my last year of college, I co-created
TextAttack, a Python
framework for adversarial attacks in NLP. See the
paper or
blog post. We also published a
paper arguing for
stricter evaluation of adversarial examples in NLP. While at Ought, I
co-created the
RAFT Benchmark.
Miscellany
Battlecode
I competed in
Battlecode for
several years, a competition for high-school and college students to
program the best bot to win a strategy game. My team advanced to the top
4 3 out of 6 years. See
the README here for more.
Clash Royale
During the end of high school and beginning of college, I played a lot of Clash Royale, a
mobile game. I did well in a few competitions and built up a bit of a following before quitting.
Speedcubing
During middle school and early in high school, I spent a lot of time
solving Rubik's cubes as fast as I could (and doing them one-handed and
blindfolded). My WCA profile is
here; I was decent but not world class.