Read more

Library › Deep Learning by Hand ✍️ 2026

1. Linear Algebra

2. Neural Network

3. MLP

4. SVM

5. Autoencoder

Outline

1. Encoder / Decoder (00:00)

2. Example 1 (02:58)

3. Example 2 (06:44)

4. 4 - 2 - 4 (10:46)

5. 6 - 4 - 2 - 4 - 6 (14:03)

6. L2 Loss (15:58)

7. L2 Loss Gradient (19:55)

8. Backpropagation (23:46)

9. Gradient Descent (41:07)

Recording & Workbook

Become a paid subscriber to access the full recording and the associated Excel workbook.

Read more

Library › Deep Learning by Hand ✍️ 2026

1. Linear Algebra

2. Neural Network

3. MLP

4. SVM

5. Autoencoder

Outline

1. Story (00:00)

2. Linear Kernel (05:04)

3. Polynomial Kernel (08:23)

4. Combine Linear Kernels to an SVM (11:03)

5. Combine Polynomial Kernels to an SVM (21:36)

6. Radial Basis Function (RBF) Kernel (28:05)

7. Combine RBF Kernels to an SVM (35:20)

8. Linear SVM Matrix Representation (40:35)

9. Practice (45:53)

Recording & Workbook

Become a paid subscriber to access the full recording and the associated Excel workbook.

Read more

Library › Deep Learning by Hand ✍️ 2026

1. Linear Algebra

2. Neural Network

3. MLP

4. SVM

5. Autoencoder

Outline

1. MLP (3 → 4 → 3 → 2 → 1) (00:26)

2. PyTorch to Matrices (07:57)

3. Output Layer for Prediction (16:55)

4. Output Layer for Classification (18:19)

5. Softmax (19:16)

6. Softmax (numerical stability) (27:59)

7. Batch Normalization (30:32)

8. Layer Normalization (37:50)

9. Practice (39:52)

Recording & Workbook

Become a paid subscriber to access the full recording and the associated Excel workbook.

Read more

Library › Deep Learning by Hand ✍️ 2026

1. Linear Algebra

2. Neural Network

3. MLP

4. SVM

5. Autoencoder

Outline

1. PyTorch to Graph to Matrix (00:00)

2. Matrix to Graph (08:38)

3. Bias (13:16)

4. Batch (16:31)

5. ReLU Activation (20:53)

6. Hidden Layer (26:05)

7. Sigmoid Activation (28:01)

8. Visualize (2-2-1) (32:04)

9. Practice (45:53)

Recording & Workbook

Become a paid subscriber to access the full recording and the associated Excel workbook.

Read more

Library › Deep Learning by Hand ✍️ 2026

1. Linear Algebra

2. Neural Network

3. MLP

4. SVM

5. Autoencoder

Outline

1. Dot Product (00:00)

2. Matrix Multiplication (04:47)

3. Broadcast (10:39)

4. Random (14:27)

Recording & Workbook

Become a paid subscriber to access the full recording and the associated Excel workbook.

Read more

This Week’s Seminar

Title: Nuro’s autonomous AI hill-climb Lessons of deploying 50+ AI Employees and Autoresearch agents for self-driving car

Bio: Richard Hsu leads Nuro’s Frontier AI Engineering team, where his team drove Nuro’s rapid AI ascent through three generations of work: from the adhoc AI chatbot you prompt, to the AI Employee that solves problems on its own, to the continuous self-improving autoresearch AI agent infrastructure. Each infrastructure and use case is shipped in production with real ROI and learnings. And this talk is the story of the ascent.

Reading: Onboarding AI Employees at Nuro

Date: July 16th, 2026

Time: 2pm (Pacific Time)

Register: https://luma.com/odzruz0q

Previous Seminars

1. Daniel Svonava, Superlinked

• Why did you shrink your team from 25 engineers to 12?

• Why did shipping more code let him run the team with fewer engineers, not more?

• How do you review code when there’s far too much of it to read?

4. Patrick Devine, Ollama

• What is the story behind your invention of Ollama?

• How did adding support for a new model go from weeks of work to an afternoon?

• What happens to open source when people contribute code they don’t understand?

5. Kalpit Singh, SingleStore

• How does your company ‘s engineering team adopt AI?

• What actually separates an AI “copilot” from an AI “colleague”?

• If you’ve stopped reviewing AI-generated code by hand, what keeps it safe to ship?

6. Jackie Lee, Science VR

• How can you get an AI agent to build a quantum circuit that multiplies matrices?

• How do you make an AI agent’s work something you can verify, not just trust?

• Does boxing an agent in with formal logic kill its creativity?

Register: https://luma.com/odzruz0q

Library › Seminar Series 2026

1. Manifold-Constrained Hyper Connections (mHC) from DeepSeek (Jan 9, 2026)

2. How Small Models Learn Tool Use (Jan 11, 2026)

3. Generative AI (Jan 11, 2026)

4. Attention (Jan 15, 2026)

5. Google Ironwood TPU: From Bits to HBM (Jan 19, 2026)

6. 9 AI Eval Formulas You Must Know (Jan 24, 2026)

7. Meta Superintelligence Labs vs Facebook AI Research (Jan 30, 2026)

8. Transformer: Six Levels of Understanding (Feb 13, 2026)

9. OpenClaw Seminar (Feb 23, 2026)

10. PPO → DPO → GRPO → Rubrics (Mar 2, 2026)

11. Gemma 4 (May 7, 2026)

12. Qwen 3.6 (May 21, 2026)

Thank you to everyone who came out for the Frontier AI Seminar on Qwen 3.6. There is something special about a room full of people willing to slow down and trace a single token by hand, and I am grateful you spent that hour with me.

In the Gemma 4 seminar, we looked at the transformer techniques that have settled, the parts of the modern recipe you can now expect almost everywhere. In this session we turned to the parts that are still moving, and we followed a single token through three of them.

We started with position. I worked RoPE by hand, rotating the query and key vectors by their position, and then YaRN, which stretches a trained context window toward 256K or even 1M without retraining.

We then looked at gated attention. I showed the learned gate that Qwen places on the attention path, and we compared the same attention output with the gate and without it, so you could see what it actually buys over a plain GQA or MLA setup.

We finished with decode. I traced speculative decoding, where a small drafter proposes tokens and the model verifies them in parallel, and then multi-token prediction, where the model emits several tokens in one step instead of one at a time.

Here is what I wanted you to walk away with. Once you have rotated the vector, gated the output, and drafted a token by hand, the acronyms stop being intimidating, and you can look at any model's spec sheet, its context length, its attention variant, its decode speed, and see what is actually doing the work.

Outline

1. PE, RoPE, YaRN

   • No Positional Encoding

   • Linear Positional Encoding

   • Sinusoidal Positional Encoding

   • Input vs Key Positional Encoding

   • RoPE: Rotary Positional Encoding

   • YaRN: Yet another RoPE extensioN

2. Gated Attention

3. Decoding: Fast and Furious

   • Auto-Regressive

   • Train

   • Speculative Decoding

   • Multi-Token Prediction

Recording & Workbook

Limited-time free: the full recording and the Excel workbook are open to everyone for now.

⬇ Excel Workbook

Each post shrinks one model down to an example that fits on one page, and works it end to end by hand, so you can follow every calculation with a pen. The recurring revelation is that, underneath the jargon, almost every model is just a series of matrix multiplications.

I have since revised the whole series and reordered it from the simplest building block to the most advanced. I hand-calculated every example myself; on a few occasions, sharp-eyed readers caught my mistakes, and their corrections are noted in the posts.

Start at the top, or jump to whichever model you have always wanted to understand. Grab a pen. ✍️

1. Matrix Multiplication by Hand ✍️

2. Multi Layer Perceptron (MLP) by Hand ✍️

3. Backpropagation by Hand ✍️

4. SVM by Hand ✍️

5. Batch Normalization by Hand ✍️

6. Dropout by Hand ✍️

7. Recurrent Neural Network (RNN) by Hand ✍️

8. LSTM by Hand ✍️

9. Deep RNN by Hand ✍️

10. Self Attention by Hand ✍️

11. Transformer by Hand ✍️

12. Autoencoder by Hand ✍️

13. Variational Auto Encoder (VAE) by Hand ✍️

14. Sparse Auto Encoder (SAE) by Hand ✍️

15. Generative Adversarial Network (GAN) by Hand ✍️

16. Sampling a Sentence by Hand ✍️

17. Residual Network by Hand ✍️

18. U-Net by Hand ✍️

19. Discrete Fourier Transform by Hand ✍️

20. Graph Convolutional Network (GCN) by Hand ✍️

21. CLIP by Hand ✍️

22. Vector Database by Hand ✍️

23. Mixture of Experts (MoE) by Hand ✍️

24. Switch Transformer by Hand ✍️

25. Mamba's S6 by Hand ✍️

26. Sora's Diffusion Transformer (DiT) by Hand ✍️

27. BitNet by Hand ✍️

28. Reinforcement Learning with Human Feedback (RLHF) by Hand ✍️

Read more

Read more

Matrix multiplication felt abstract…

Read more

Models

Attention

Fine-Tuning

Math

Activation

Matrix Multiplication

Calculating AI by Hand ✍️

Videos

Seminar Series 2026

Seminar Series 2025

Deep Learning by Hand ✍️ 2026

Drawings

Frontier AI Drawings

Workbooks

Token Problems

What I cannot create, I do not understand.

— Richard Feynman

💡 Pro tip: Get reimbursed

AI by Hand membership is designed for professionals working in industry. Most members don't pay out of pocket. Their companies cover it. The membership fits naturally within your professional development budget, and it costs less than a single traditional training course. Just ask.

📧 Template email to your manager/HR:

Hi [Manager's Name],

I'd like to join AI by Hand ($29/month), the research publication of By Hand Research founded by Prof. Tom Yeh.

Many leading AI companies already sponsor their teams because it's the only place that systematically connects foundational math and algorithms to frontier architectures and grounds it in actual interpretability research. The material maps directly to the AI models we use in practice, making it immediately relevant to my current projects.

Can we expense this under our training / professional development budget?

Thanks!

Read more

This is not the first time I have tried to make you pick up a pen and write on paper. In my previous Agentic AI online seminars, I asked you to sketch a diagram, draw a line, and many of you did. This time I want to bring it up a notch: can I get you to do some math? Don't worry, simple math: adding numbers, multiplying them, working things in reverse. The goal is to build real intuition about how agents work by solving problems by hand, rather than passively reading or watching.

This first workbook is about tokens. The next one will cover the context window.

Now let's see if you can solve all 20 problems.

1. Count the Tokens

2. Subword Splitting

3. Punctuation Counts

4. Tokens per Word

5. Will It Fit?

6. The Price of a Call

7. Two Kinds of Token

8. Input or Output?

9. The Two-Part Bill

10. End-to-End Call Cost

11. Comparing Three Calls

12. The Average Call

13. The System Prompt

14. The System Prompt Tax

15. Calls in a Budget

16. Tokens on a Budget

17. How Long Will It Take?

18. The Split Matters

19. Pricing per Million Tokens

20. Tokens from the Bill

Library › Token Problems

1. Count the Tokens

2. Subword Splitting

3. Punctuation Counts

4. Tokens per Word

5. Will It Fit?

6. The Price of a Call

7. Two Kinds of Token

8. Input or Output?

9. The Two-Part Bill

10. End-to-End Call Cost

11. Compari…

Read more

Library › Token Problems

1. Count the Tokens

2. Subword Splitting

3. Punctuation Counts

4. Tokens per Word

5. Will It Fit?

6. The Price of a Call

7. Two Kinds of Token

8. Input or Output?

9. The Two-Part Bill

10. End-to-End Call Cost

11. Compari…

Read more

Library › Token Problems

1. Count the Tokens

2. Subword Splitting

3. Punctuation Counts

4. Tokens per Word

5. Will It Fit?

6. The Price of a Call

7. Two Kinds of Token

8. Input or Output?

9. The Two-Part Bill

10. End-to-End Call Cost

11. Compari…

Read more

Library › Token Problems

1. Count the Tokens

2. Subword Splitting

3. Punctuation Counts

4. Tokens per Word

5. Will It Fit?

6. The Price of a Call

7. Two Kinds of Token

8. Input or Output?

9. The Two-Part Bill

10. End-to-End Call Cost

11. Compari…

Read more

Library › Token Problems

1. Count the Tokens

2. Subword Splitting

3. Punctuation Counts

4. Tokens per Word

5. Will It Fit?

6. The Price of a Call

7. Two Kinds of Token

8. Input or Output?

9. The Two-Part Bill

10. End-to-End Call Cost

11. Compari…

Read more