Caleb Maresca

Hi, I'm Caleb! I'm a PhD student at New York University with a focus on the intersection of artificial intelligence and economic systems. My research interests span many areas, including macroeconomic modeling of the effects of AI, laboratory experiments evaluating the strategic behavior of AI agents, ML for causal inference, as well as AI safety and mechanistic interpretability.

Currently, I'm applying these insights as a Research Scientist Intern at Upstart, where I work on causal machine learning methods for real-world applications.

Résumé

AI & Machine Learning Research

Strategic Wealth Accumulation Under Transformative AI Expectations

The rapid progress of development in the field of artificial intelligence may profoundly reshape the global economy by both increasing productivity and automating away many jobs. This paper explores how households adjust their economic behavior today in anticipation of transformative AI (TAI). Building on previous research, I introduce a novel mechanism where the future reallocation of labor from humans to AI systems owned by wealthy households creates a zero-sum contest for control over AI resources—driving changes in current savings decisions and asset prices.

Paper Code

SenGen: Scenario Generator for LLM RL Agents

SenGen is a Python package designed to generate interactive scenarios for training and evaluating LLM-based reinforcement learning agents, with a particular focus on ethical decision making and AI safety.

Code

Monet: Mixture of Monosemantic Experts for Transformers Explained

MONET is a novel neural network architecture using extreme expert specialization (~250k experts/layer) to improve interpretability without sacrificing performance. I explain how it works, present a new interpretation of the architecture, and propose efficiency improvements.

Read on LessWrong

NSCAN: News-Stock Cross-Attention Network (with Nishant Asati)

NSCAN is a novel deep learning model for predicting multiple stock returns simultaneously using financial news data and cross-attention mechanisms. Unlike traditional approaches that predict returns for individual stocks in isolation, our model captures cross-asset relationships and market interactions.

Code

Other Academic Research

The (In)Effectiveness of State R&D Grants

States match federal SBIR grants to incentivize local R&D, but do they work? Looking at Kentucky's aggressive state match program using synthetic control methods, I find mixed results, including a large yet statistically insignificant increase in private R&D, and a surprising negative effect on new business formation.

PDF

Happier than Thou: Causal Evidence for the Effect of Religion on Subjective Well-Being (with Joseph Lee)

Previous research shows religious people are happier, but correlation isn't causation. Using novel econometric techniques and World Values Survey data, I show both believing in God and attending religious services actually make people happier across seven countries.

PDF

Cognitive Biases are Critical in Conflict Bargaining

Cognitive biases fundamentally alter how negotiators behave in conflicts. Unlike standard bargaining theory which predicts costly conflicts should never occur, prospect theory shows that when parties have conflicting reference points, fighting can become unavoidable.

PDF