Haoran Zheng
AI Researcher
Pre-doc Fellow at The University of Chicago Booth in Quantitative Marketing with Prof. Andreas Kraft. I'm applying to PhD programs in Computer Science, Marketing, or Cognitive Science this cycle!
What guides my research
No two intelligences are exactly alike, but they do often rhyme.
My work focuses on improving language models by drawing inspiration from human cognition, and on how language models, in turn, shape the way humans perceive the world. It started in AI InterpretabilitySCENE: an empirical evaluation framework for explainable AI techniques in NLP tasks, using soft counterfactual explanations. and AI ConsciousnessConsciousnessBench: a benchmark for evaluating consciousness-relevant traits in large language models through theory-grounded self-reports.. Most recently, I designed AMORA hybrid architecture inspired by Daniel Kahneman's Thinking, Fast and Slow — appending a few slow, deliberate attention layers gated by the model entropy to a fast, intuitive recurrent model. and ConceptizerA training framework that extends next-token prediction with a reward for meaning — so language models learn to predict meaning, not just the next token. (advised by Dr. Chen Shani – Stanford). More on my Google Scholar.
In past lives, I did my Master's in Data Science at The University of Chicago (GPA 4.0), triple-majored in Economics, Finance, and History at The Ohio State University (top 1% Honors), worked as a research assistant in Behavioral Science (Dr. Nick Epley – UChicago) and in History (Dr. Christopher A. Reed – OSU), served as a finance analyst at NASDAQ-listed CAAS Inc., and taught Statistics at Ohio State.
