Andrew Jun L.

PhD Candidate in Cognitive Science @ UCLA

United States

About

I love understanding and explaining things clearly! At UCLA, I work on AI interpretability and cognitive science (though I actually think that AI interpretability is a field under the cognitive sciences!). Previously, I studied philosophy & cognitive science at Williams College and Oxford University (study-abroad). Google Scholar: https://scholar.google.com/citations?user=XuzjypEAAAAJ&hl=en

Experience

  • Research Scholar at MATS Research
    Jun 2026 - Present · 2 mos

    Mentors: Paul Reichers, Adam Shai at Simplex - MATS 10.0 scholar

  • Research Scholar at Blavatnik School of Computer Science and AI, Tel Aviv University
    Feb 2026 - Present · 6 mos

    Mentor: Mor Geva - Studying how LLMs learn how to do a task from few examples (a.k.a., in-context learning)

  • Research Fellow at SPAR Research
    Feb 2026 - Present · 6 mos

    Mentor: Xavier Poncini at Simplex - Studying whether LLMs know and reason about the state of the world as it processes a user's prompt, using HMMs as a mathematically tractable case-study of world states.

  • UCLA (3 yrs 11 mos)
    • Graduate Research Assistant - Co-Mind Lab
      Sep 2023 - Present · 2 yrs 11 mos

      Mentors: Rick Dale, Hongjing Lu, Alexia Galati - Furthered theory of interpersonal dynamics by developing a Bayesian dynamical system model explaining joint action/coordination.

    • Graduate Research Assistant - Computational Vision and Learning Lab
      Sep 2022 - Present · 3 yrs 11 mos

      Mentors: Hongjing Lu, Taylor Webb, Keith Holyoak - Improved accuracy by 10% on relation-based few-shot image classification using a graph-matching model. - Analyzed how LLMs perform relational reasoning by identifying attention circuits within the models. - Identified systematic failures of multimodal LLMs in classifying objects into correct relational roles, motivating graph-based models.

    • Graduate Research Assistant - Reasoning Lab
      Sep 2022 - Present · 3 yrs 11 mos

      Mentors: Keith Holyoak, Hongjing Lu - Identified sources of generalizability in CNN representations trained on relation-based image classification. - Wrote paper arguing that analogical mapping results in generalizable human mental representations of relation-based concepts.

  • Machine Learning Intern at Wright-Patterson Air Force Base
    Jun 2024 - Feb 2026 · 1 yr 9 mos

    Mentors: Emily Conway, Jeremy Gwinnup, Grant Erdmann, Eric Hansen, Tim Anderson - Reduced video captioning error of video LLMs and scene-graph transformers by training on object- and relation-based captions. - Developed NLP pipeline to identify news trends of objects, relations, and semantic triplets (object-relation-object) in captions.