Alessandro Morosini

Alessandro Morosini

I am a PhD student at the MIT Operations Research Center (ORC), where I am fortunate to be advised by Professor Georgia Perakis and Professor Chara Podimata. I previously graduated from the Master of Business Analytics program at MIT ('25) and hold a Bachelor's in Economics and Computer Science from Bocconi University ('24). Outside of research, you'll usually find me playing football (soccer, for my American friends) or training outdoors.

Research interests

I am broadly interested in AI reliability and auditing. In particular, I study how and why LLMs (over)condition on context, and how this affects their behavior. For example, I am interested in multi-turn conversations, where models over-index on earlier (possibly wrong) turns. Or over-personalization, where models overly adapt their answers to (possibly irrelevant) cues about the user's identity or beliefs. Or AI persona simulations, where models over-index on (possibly stereotypical) stated features of the persona to be simulated. Besides trying to fix these problems, I have also focused on auditing them (with a focus on political information), e.g., by auditing LLM answers to election questions or biases in the political content recommender systems serve. I am also interested in uncertainty quantification, particularly how models express confidence in their answers and how we can use this to make these systems more reliable.

Research

(αβ) denotes alphabetical author order; * denotes equal contribution.