Lexin Zhou

I’m a CS PhD candidate at Princeton University, advised by Peter Henderson and closely collaborate with Tom Griffiths. I love both CS and Cognitive Science. I work on data, evals and continual learning of language models and agents. A few technical directions that I’m recently excited about:

  • Efficient, holistic evals methodologies that inform model capabilities and limits
  • Generalizable optimization metrics that fight against Goodhart’s Law
  • Closed loops of data valuation and targeted synthetic data generation

I’ve spent time in research/consultancy roles at MSR, Google DeepMind, OpenAI, Meta AI, and European Commission. My work has been featured in Nature, Financial Times, Microsoft Research, MIT Tech Review, Forbes, IEEE Spectrum, El País, New Scientists, QbitAI, IBM, among others. I am grateful to have my latest and ongoing projects being generously supported by OpenAI, Google DeepMind and MSR.

I’m always open to interesting conversations. Feel free to reach out via email or Twitter.

selected publications

  1. NATURE
    General Scales Unlock AI Evaluation with Explanatory and Predictive Power
    Lexin Zhou, Lorenzo Pacchiardi, Fernando Martı́nez-Plumed, Katherine M. Collins, Yael Moros-Daval, Seraphina Zhang, and 20 more authors
    Nature, 2025
    PDF Code
  2. NATURE
    Larger and More Instructable Language Models Become Less Reliable
    Lexin Zhou, Wout Schellaert, Fernando Martı́nez-Plumed, Yael Moros-Daval, Cèsar Ferri, and José Hernández-Orallo
    Nature, 2024
    🌟 Extensive Media Coverage PDF Code