Skip to content
View max-ng's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report max-ng

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
max-ng/README.md

Hi there 👋

One day, humans will build computers near the Sun, power them with a Dyson sphere, and wait for the results near a black hole. Relativistic time dilation could let us observe far more computation than our local time would normally allow, bringing us closer to the experience of near-infinite compute.

Without computers, I would not have what I have today.
I am grateful for computing, and I want to give something back to the field.


Some of my theoretical work and technical notes

  • A Mathematical Framework for Intelligence
    [arXiv] (NEW)

  • On the Definition of Intelligence
    [arXiv] (NEW) In Artificial General Intelligence, Springer Lecture Notes in Computer Science.

  • 100,000 simulations for an interesting Monte Carlo problem
    [url]


Some of my engineering work

  • Looped Neural Network
    [arXiv]
    The first latent reasoning paper, introducing a looped architecture that shows large performance gains for GPT-style models and natural language tasks.

  • Self Meta Pseudo Labels: Meta Pseudo Labels Without The Teacher
    [arXiv]
    Single-model semi-supervised learning, where one model acts as both teacher and student. State-of-the-art on CIFAR at the time.

  • Handcrafted a GRU from scratch in 2016, including backpropagation through time, while I was first teaching myself to program. [code]

  • The first multilingual English-Chinese LLaMA fine-tuned model in the world. [code]

  • Fastest multiple pickup-dropoff N-to-N routing engine
    [closed]


Some of my product work

  • Turn Claude, Codex, Cursor, Openclaw, and agent sessions into encrypted data assets for future AI employees. [code] (NEW)

Contact

Contact email

Pinned Loading

  1. datamoat datamoat Public

    Export, back up, search, analyze, and reuse ChatGPT / Claude / Codex / Cursor / DeepSeek / Qwen data + skills + attachments locally

    TypeScript 208 10

  2. recurser recurser Public

    Reduce VRAM usage on transformer models

    Python 9

  3. megaboost megaboost Public

    Semi-supervised AutoML library on top of PyTorch.

    Python 51 1

  4. GRU-recurrent-network GRU-recurrent-network Public

    GRU recurrent network in javascript. GRU is a powerful model in sequential modeling. you may stack it to do some interesting things( generate articles, musics, etc). http://arxiv.org/pdf/1412.3555v…

    JavaScript

  5. THE-DEFINITION-OF-FENG-A-NEW-MEASURE-OF-RESOURCE-DISTRIBUTION THE-DEFINITION-OF-FENG-A-NEW-MEASURE-OF-RESOURCE-DISTRIBUTION Public

  6. pig_zero pig_zero Public

    fine-tuned LLaMA model on both English and Chinese

    4