Wenyu Du

Wenyu Du

Co-Founder

DualverseAI

Biography

I am currently a Co-Founder at DualverseAI. Our goal is to advance AI-driven scientific discovery by building an open-world multi-agent environment for AIs. We believe that, rather than relying on human-guided structural design, scientific discoveries will naturally emerge through rich contextual interactions and long-term exploration among agents. We developed The Station, which achieves strong results on Math, ML and biology benchmarks.

We are looking for collaborators and interns! Interested parties can email info@dualverse.ai or fill out this form.

I received my PhD in Computer Science from the University of Hong Kong (HKU) in 2025, advised by Professor Reynold Cheng. Prior to that, I obtained my MSc and BSc from the University of Southampton and University of Nottingham, respectively.

My research interests include AI-Driven Scientific Discovery, Mechanistic Interpretability, and the efficient training of large models.

Interests

  • AI-Driven Scientific Discovery
  • Mechanistic Interpretability
  • Efficient LLM Training

Education

  • PhD in Computer Science

    University of Hong Kong

    Mar. 2021 - Nov. 2025

  • MSc in Data Science

    University of Southampton

    Sep. 2016 - Nov. 2017

  • BSc in Computer Science

    University of Nottingham

    Sep. 2011 - July 2016

Experience

  • Co-Founder

    DualverseAI

    Nov. 2025 - Present

    Building "The Station", a generative simulation platform for AI-driven scientific discovery.

  • Research Intern

    NVIDIA

    Jun. 2024 - Aug. 2024

    Project: Unlocking continual learning ability in language models (accepted by EMNLP 2024).

Selected Recent Publications

* = equal contribution


Thinker: Learning to Think Fast and Slow.
The Thirty-Nineth Annual Conference on Neural Information Processing Systems (NeurIPS, Poster), 2025.

Stacking Your Transformers: A Closer Look at Model Growth for Efficient LLM Pre-Training.
The Thirty-Eighth Annual Conference on Neural Information Processing Systems (NeurIPS Spotlight), 2024.

Unlocking Continual Learning Abilities in Language Models.
The 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP Findings), 2024.

Contact

  • wenyudu@yahoo.com
  • Hong Kong