AI by Hand ✍️

AI by Hand ✍️

Scaling LLMs by Hand ✍️ in JAX (Aug 13, 2025)

AI by Hand ✍️ Seminars

Prof. Tom Yeh's avatar
Prof. Tom Yeh
Aug 13, 2025
∙ Paid

Library › Seminar Series 2025

  1. Draw DeepSeek (Feb 20, 2025)

  2. LLaMA 1 → 2 → 3 → 4 (May 14, 2025)

  3. Scaling LLMs by Hand ✍️ in JAX (Aug 13, 2025)

  4. GPU by Hand ✍️ Together (Nov 2, 2025)

  5. Agent Hallucination (Nov 12, 2025)

  6. Gated Attention ~ NeurIPS 2025 Best Paper (Dec 18, 2025)

When Google’s JAX team first approached me to sponsor a special lecture, I told them, “You asked the wrong person. I’ve never written a single line of JAX code, but I can walk people through the math and algorithms behind the syntax. Is that okay?”

They said, “Absolutely. That’s exactly what we want you to do.”

In preparing for this lecture, I got to meet many people on Google’s JAX team, including the creators. At one JAX event, Roy Frostig, one of JAX’s creators, opened his keynote with a single slide: “Matrix Multiplication is Everything!”

If you know me, you know I couldn’t agree more.

In this seminar, I walk you through:

  • How attention scales to millions of tokens

  • What vectorization really means in JAX

  • How to understand @vmap (vectorization map)

All by hand ✍️ in Excel, so you can see exactly what’s going on inside the black box.

P.S. This special lecture is sponsored by Google. If you’d like to sponsor a future AI by Hand ✍️ lecture to teach an important topic to a large, global AI audience, feel free to reach out.

Feedback

Recording

Become a paid subscriber to access the full recording.

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2026 Tom Yeh · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture