Alek Dimitriev

I’m a machine learning engineer at Google, working on Gemini finetuning and inference for teams across Alphabet, including a Nobel prize winner! I’ve been fortunate to work on a wide range of problems, ranging from multimodal capabilities augmentation, to both generative and non-generative tasks like text embedding, to automating model hyperparameters and sharding to improve training and inference efficiency.

In my free time I like to read mostly non-fiction, and on rare occassions interview people about their work. You can watch my interview with Scott Aaronson on LLM watermarking, or read my interview with Richard Dawkins on the future of religion & skepticism.

Before Google, I did my PhD in machine learning at the University of Texas at Austin, advised by Prof. Mingyuan Zhou, where my dissertation was focused on gradient estimation for discrete variables via dependent Monte Carlo samples. I also hold an M.Sc. and B.Sc. in computer science from the University of Ljubljana.

Selected papers:

  1. Gemini 1.5: Unlocking Multimodal Understanding Across Millions of Tokens of Context
    Gemini Team
    Technical Report, Feb 2024
  2. Gemini: A Family of Highly Capable Multimodal Models
    Gemini Team
    Technical Report, Dec 2023
  3. CARMS: Categorical-Antithetic-REINFORCE Multi-Sample Gradient Estimator
    Alek Dimitriev and Mingyuan Zhou
    NeurIPS: Neural Information Processing Systems, Dec 2021
  4. ARMS: Antithetic-REINFORCE-Multi-Sample Gradient for Binary Variables
    Alek Dimitriev and Mingyuan Zhou
    ICML: International Conference on Machine Learning, Jul 2021