This was a talk I gave at the AI for Peace Workshop at ICLR 2026. The whole workshop was amazing and I want to give so much thanks to the organizers and other speakers. Having these kinds of discussions, especially at a large ML conference like ICLR, is super important and I hope there is a lot more of this in the future!
https://aiforpeaceworkshop.github.io
This talk summarizes a lot of thinking and investigation I did in the aftermath of leaving AI research after YOLOv3. None of these ideas are original or unique I’m just approximately parroting the thoughts of much smarter people. There’s a lot of topics I touch on very lightly that are worth investigating more. If you’re interested here’s some starting points I’d recommend:
The slides can be found here: https://docs.google.com/presentation/d/1WRA6e4Qj3Iqsj1oj2SUZdOSnVUYv5JUweqO886scOhw/edit?usp=sharing
I’m working on the OlmoEarth team at Ai2 building state-of-the-art AI for environmental and humanitarian groups.
Platform: OlmoEarth Platform
Project blog: OlmoEarth
Technical blog: OlmoEarth Models
Paper (CVPR 2026): OlmoEarth
Github: olmoearth_pretrain
I am a research scientist at Ai2 working on machine learning for the environment.
I was a graduate student advised by Ali Farhadi. I work on computer vision.
I maintain the Darknet Neural Network Framework, a primer on tactics in Coq, occasionally work on research, and try to stay off twitter.
Outside of computer science, I enjoy skiing, hiking, rock climbing, and playing with my Alaskan malamute puppy, Kelp!
The Real AI Alignment Problem - AI for Peace Workshop, ICLR 2026
The Ancient Secrets of Computer Vision - University of Washington, Spring 2018
Computers can see. Now what? - TEDxGateway in Mumbai, India
How computers learn to recognize objects instantly - TED2017 in Vancouver, Canada
ECCV 2016
CVPR 2016, OpenCV People's Choice Award
... Either The Least Or Most Employable Person Ever
He knows how to do the internet.
I think the little pony resume is probably not the right choice for grad school.