Research

In academia, I’m working on the computational complexity of bounded arithmetics supervised by Albert Atserias, although soon I’ll begin applying tools from computational complexity to Machine Learning robustness. I’m also working with Juan Aguilera on higher-order computability and non-classical logics.

At the Center on Long-term Risk, I’m using mathematical modelling to understand machine-augmented power conflict, and build robust game-theoretic mechanisms for cooperation. I’m also working on AI-relevant algorithmic implementations of Decision Theory with Abram Demski (Machine Intelligence Research Institute). At MATS I pursued conceptual and mathematical research on safety improvements to Language Model training.

Publications

Type-2 Feedback Computability. Juan Aguilera, Martín Soto. Submitted for publication

Logically Updateless Decision-Making. Martín Soto, Abram Demski, Alex Appel. Technical report, PIBBSS Symposium 2023. Working paper.

The real numbers in inner models of set theory. Martín Soto. Bachelor’s Thesis, arXiv preprint (arXiv:2206.10754), and Submitted for publication.

Network-Iterated Prisoner’s Dilemma. Martín Soto. Bachelor’s Thesis, arXiv preprint (arXiv:2211.16329).

The marginal impact of cooperation-inducing interventions. Martín Soto, Tristan Cook. Working paper.

Talks

Open-source Game Theory, in Logic Seminar Barcelona 2023.

How can we effectively govern artificial intelligence?, in Adevinta Barcelona 2023 for Effective Altruism Barcelona.

Logic & Music, in Music and Engineering 2022, invited speaker at the Polytechnic University of Barcelona.

Music & Mathematics, in Music and Engineering 2020, invited panelist at the Polytechnic University of Barcelona.