Andrew Xu

Hi, I’m Andrew, a junior at Yale studying Computer Science and Mathematics. I’m interested in how language models reason – and whether we can do better than the autoregressive paradigm that dominates today.

Currently, I’m investigating coarse-to-fine summarization as a mechanism for improving reasoning in diffusion language models, working with Arman Cohan (Yale NLP) and Neil Band (Stanford NLP). I’m also integrating SDPO, a new RL framework for LLM post-training, with veRL and scaling it for multi-node training — along with exploring extensions to the algorithm itself. Previously, I published work at ICML 2024 on algorithmic fairness: specifically, designing selection algorithms that achieve near-optimal utility while correcting for systematic bias against underrepresented groups in multi-institutional matching settings. This summer, I’ll be at Jane Street as a Quantitative Trading intern.

Outside of research, I sing with Doox of Yale and the Yale Glee Club, and I’ve recently started to read more novels. I think a lot about stories – across writing, film, and music – and what they reveal about humanity.

I’d love to talk about advancing the frontiers of language modeling, AI infrastructure, and the ethics of how we build and deploy AI systems. And if you have a novel or film that changed how you see something, I want to hear about it – reach out to me using the email button.

Responsibility And The Adoption Of Safe Artificial Intelligence Systems

1 Introduction The promise of artificial intelligence (AI) has brought reverberating impacts throughout aspects of everyday life, from chatbots to education to transportation, but nowhere has it been felt as decisively—or controversially—as in warfare. The advancement of Lethal Autonomous Weapons Systems (LAWS) has been a strategic goal of militaries for decades, driven by the hope of deploying agents that are more agile, precise, and capable than human soldiers, all while removing human combatants from immediate physical risk. However, the ethical implications of delegating the use of lethal force to non-human entities are frequently called into question. ...

Large Language Model Post-Training as a Form of Socialization

1 Introduction Over the past decade, natural language processing has evolved from statistical prediction to deep learning systems that emulate complex, human-like behaviors. The invention of the Transformer architecture (Vaswani et al., 2017) has allowed the scale of training necessary for large language models, and new research on the technology has developed models that can reason, possess world knowledge, and be helpful chatbots. While architecture and scale have been major drivers of progress, many of the most interesting behavioral breakthroughs have occurred in post-training: the stage after self-supervised pre-training where models are refined through supervised fine-tuning, reinforcement learning from human feedback (RLHF), and other alignment procedures. The base objective of next-token prediction is a proxy for language understanding and world knowledge, but it is agnostic to conceptions of “truth”1 or ethical norms. Post-training injects human preference data into these systems, teaching them to not merely predict text, but to perform appropriately within human social and moral contexts. The post-training process of an LLM resembles socialization in human development: the transformation of raw cognitive potential into an assistant capable of following norms, persuading others, and explaining its own reasoning. Yet this human-like training also introduces epistemological hazards. Models are optimized for approval, not “truth”—a tension that mirrors classic psychological findings about human introspection and reasoning. ...

Centralized Selection with Preferences in the Presence of Biases

Authors L. Elisa Celis, Amit Kumar, Nisheeth K. Vishnoi, Andrew Xu Links arXiv / code / slides Abstract This paper considers the scenario in which there are multiple institutions, each with a limited capacity for candidates, and candidates, each with preferences over the institutions. A central entity evaluates the utility of each candidate to the institutions, and the goal is to select candidates for each institution in a way that maximizes utility while also considering the candidates’ preferences. The paper focuses on the setting in which candidates are divided into multiple groups and the observed utilities of candidates in some groups are biased–systematically lower than their true utilities. The first result is that, in these biased settings, prior algorithms can lead to selections with sub-optimal true utility and significant discrepancies in the fraction of candidates from each group that get their preferred choices. Subsequently, an algorithm is presented along with proof that it produces selections that achieve near-optimal group fairness with respect to preferences while also nearly maximizing the true utility under distributional assumptions. Further, extensive empirical validation of these results in real-world and synthetic settings, in which the distributional assumptions may not hold, are presented. ...