Max Gupta
Psychology × Computer Science
I’m an AI researcher in stealth based in Cambridge, MA, where I work on steering and evaluating language models through the lens of cognitive science — in particular distillation: approaches for fine-tuning representations and concepts from cognitive science into language models.
Before that I obtained my master’s at Princeton, in the Computational Cognitive Science lab and the Princeton Laboratory for Artificial Intelligence with Tom Griffiths, working on meta-learning, few-shot concept learning, and human inductive biases in neural networks (see thesis). Earlier I spent three years as a consultant and software engineer, and studied applied mathematics at Columbia University.
Education
- Princeton University
- M.S.E., Computer Science. 2024–2026. Graduate Fellowship (full tuition and teaching stipend).
- Columbia University
- B.A., Applied Mathematics. 2017–2021. Dean’s List; Heinrich Research Fellowship; Spritz Family Research Grant.
Representative Publications
- Gupta, M., Shin, C., Lake, B., & Kwan, T. (2026). Modeling student arithmetic reasoning by distilling cognitive priors into language models. Under review
- Gupta, M., Campbell, D. I., & Griffiths, T. L. (2026). Meta-learning captures human-like geometric sensitivity. Proceedings of the 48th Annual Conference of the Cognitive Science Society.
- Gupta, M., Rane, S., McCoy, R. T., & Griffiths, T. L. (2025). Convolutional neural networks can (meta-)learn the same-different relation. Proceedings of the 47th Annual Conference of the Cognitive Science Society. Also presented as a poster at the Frontiers in NeuroAI Symposium, Kempner Institute, Harvard University.
- Bencomo, G., Gupta, M., Marinescu, I., McCoy, R. T., & Griffiths, T. L. (2025). Teasing apart architecture and initial weights as sources of inductive bias in neural networks. Proceedings of the 47th Annual Conference of the Cognitive Science Society.
Elsewhere
Contact
mg7411@princeton.edu · Curriculum vitae (PDF) · LinkedIn
Cambridge, Massachusetts.