David Koplow
PhD student · MIT EECS
I’m a PhD student in Electrical Engineering and Computer Science at MIT, advised by Leslie Kaelbling and Tomaso Poggio. I study how prior learning shapes new learning and how AI systems reason beyond their training distribution, drawing on neuroscience and human intelligence.
I hold an MEng in Artificial Intelligence and an SB with majors in Artificial Intelligence and Neuroscience, both from MIT.
Publications and manuscripts
Tool Use Reduces Depth-Induced Collapse in OOD Reasoning
David Koplow, Tomer Galanti, Tomaso Poggio.
Ubiquity of Emergent Hebbian Dynamics in Regularized Learning
David Koplow, Tomaso Poggio, Ziyin Liu.
Intelligent Entrainment
Selected projects
Laws of Emergence [PDF]
An essay on entropy and emergence.
Optimizing Eta Kappa Nu Tutor Assignments [PDF]
A tutor-matching algorithm that improved match rates by about 40% on historical data.
Automatic Shadow-Based Satellite Feed Verification [PDF]
Estimating satellite image acquisition times from shadows and LiDAR.
Exploration of the Graph Isomorphism Problem [PDF]
An empirical study of an O(n⁴) graph-isomorphism algorithm and its limitations.
Honors and awards
- Brain and Cognitive Sciences Excellence Award ()
- Phi Beta Kappa ()
- Neo Scholar ()
- Tau Beta Pi ()
- Eta Kappa Nu ()
- Schwarzman College of Computing Undergraduate Advisory Group ()
Some other things I’ve done
Previously, I was a founding ML engineer at Reducto AI and co-founded Tau Clinical.
I was a teaching assistant at MIT for Representation, Inference and Reasoning in AI (Spring 2026) and Quantitative Methods for Natural Language Processing (Fall 2025).
I previously served as co-president of AI@MIT and MIT’s Eta Kappa Nu chapter.
I co-authored the “GPT-4 Has Not Aced MIT” rebuttal and gave a talk at MIT about evaluating language models in education.
In 2020, I built 140+ interactive JavaScript simulations in physics, biology, and chemistry for Stem Development’s digital textbooks, used by tens of thousands of high school students.