Marina Mancoridis

I am a second-year Computer Science PhD student at MIT, advised by Sendhil Mullainathan. I am interested in the intersection of artificial intelligence with behavioral science and decision-making.

I graduated summa cum laude from Princeton University with a degree in Computer Science. At Princeton, I worked with Tom Griffiths in the Computational Cognitive Science Lab. Upon graduating, I received the Calvin Dodd MacCracken Senior Thesis Award for the most inventive and technically accomplished senior thesis in the School of Engineering and Applied Sciences.

Our recent paper, Potemkin Understanding in Large Language Models, formalizes and benchmarks conceptual understanding in language models—work that has gained significant attention on Twitter and in the AI community.

I also enjoy singing and playing guitar. You can download my full CV here.

news

Jun 2025	We presented our work on Potemkin Understanding at ICML 2025.
Jun 2025	I was invited to give a talk at OpenAI in San Francisco about Potemkin Understanding (slides available here). Unfortunately, I was unable to deliver the talk due to illness.
Jun 2025	We released our paper on Potemkin Understanding in Large Language Models, and received significant attention on Twitter! Our work was featured by PPC Land, the Register, and presented in a livestream via Ploutos.
Sep 2024	I started my CS PhD at MIT, advised by Sendhil Mullainathan. I received the selective MIT Presidential Graduate Fellowship to support my research.
Jun 2024	I gave a talk about my formal model of science at CogSci 2024 in Rotterdam, Netherlands.
May 2024	I graduated summa cum laude from Princeton University! Upon graduation, I received the Calvin Dodd MacCracken Senior Thesis Award for the most inventive and technically accomplished senior thesis in the School of Engineering and Applied Sciences (from around 500 theses).
Jun 2023	I gave a poster presentation on individual differences in risky decision-making at CogSci 2023 in Sydney, Australia.
May 2023	My junior paper won Princeton’s Spring 2023 Outstanding Independent Work award, for the best junior research paper in the CS department.

publications

Potemkin Understanding in Large Language Models

Marina Mancoridis, Bec Weeks, Keyon Vafa, and Sendhil Mullainathan

In ICML 2025

arXiv Bib Code

@article{mancoridis2025potemkin,
  title = {Potemkin Understanding in Large Language Models},
  author = {Mancoridis, Marina and Weeks, Bec and Vafa, Keyon and Mullainathan, Sendhil},
  journal = {In ICML 2025},
  year = {2025},
  archiveprefix = {arXiv},
  primaryclass = {cs.CL},
  url = {https://arxiv.org/abs/2506.21521},
  booktitle = {In ICML 2025},
}

Publish or Perish: Simulating the impact of publication policies on science

Marina Mancoridis, Ted Sumers, and Tom Griffiths

In CogSci 2024

Awarded Bib PDF

Princeton University Calvin Dodd MacCracken award (best thesis in School of Engineering and Applied Sciences); Princeton University Senior Thesis Award (best thesis in CS department)

@inproceedings{mancoridis2024publish,
  title = {Publish or Perish: Simulating the impact of publication policies on science},
  author = {Mancoridis, Marina and Sumers, Ted and Griffiths, Tom},
  booktitle = {In CogSci 2024},
  volume = {46},
  year = {2024},
  url = {https://escholarship.org/uc/item/9cb0b1pw},
}

To each their own theory: Exploring the limits of individual differences in decisions under risk

Joshua Peterson, Marina Mancoridis, and Tom Griffiths

In CogSci 2023

Awarded Bib PDF

Princeton University Computer Science Department Outstanding Independent Work Award (best junior-year research paper in the department)

@inproceedings{peterson2023each,
  title = {To each their own theory: Exploring the limits of individual differences in decisions under risk},
  author = {Peterson, Joshua and Mancoridis, Marina and Griffiths, Tom},
  booktitle = {In CogSci 2023},
  volume = {45},
  year = {2023},
  url = {https://escholarship.org/uc/item/0nq6m9m2},
}

selected talks

Potemkin Understanding in Large Language Models
July 2025

I was invited give a talk to OpenAI’s Economic Reserach team. The talk involved how to evaluate and benchmark large language models for depth of conceptual understanding. Here are the slides.

Governance of Emerging Technology and Tech Innovations for Next-Gen Governance: Potemkin Understanding
April 2025

I gave an invited talk at Harvard’s GETTING-Plurality group (Ash Center). I talked about potemkin understanding in large language models. Here are the slides.

Publish or Perish: Simulating how Researcher and Publisher Behaviors Impact Science
July 2022

I gave a talk at the 46th Annual Meeting of the Cognitive Science Society. My talk introduced a new theoretical model that frames science as a collective information-gathering process, using a multi-armed bandit framework. Here are the slides.