GPT in Vanilla JavaScript
🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲
Tickets are ON SALE for CascadiaJS 2026 - https://cascadiajs.com/2026
🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲🌲
We'll demystify LLM internals for web developers using a real LLM implemented entirely in vanilla JavaScript that runs in the browser. No Ph.D. needed!
00:00 - Introduction & Background
00:39 - The Power of "View Source" for Learning
01:39 - Bringing "View Source" to AI: GPT-2 in the Browser
02:29 - Live Demo: Running GPT-2 Locally in JavaScript
03:53 - Why GPT-2? Model Lineage & Simplicity
04:55 - Large Language Models as Autocomplete Engines
05:43 - Tokenization Explained
07:05 - Subword Tokenization & Vocabulary Compression
07:35 - Embeddings: Mapping Words to Math
09:07 - Semantic Relationships in Embedding Space
10:43 - How Embeddings Are Learned
11:39 - Using Embeddings in the Model
12:46 - Attention Mechanism Overview
13:38 - Contextualizing Words with Attention
15:33 - The Perceptron: Making Predictions
17:20 - From Embeddings to Predicted Tokens
18:28 - Calculating Token Probabilities & Softmax
20:21 - Full Process Recap
21:12 - Learning AI: Mastery is Within Reach
22:03 - "View Source" Principles for AI Workflows
22:50 - Demo: Spreadsheets Are All You Need Notebooks
24:48 - Resources & Getting Started