Understand how LLMs actually work.
Free, interactive lessons that build your intuition for large language models from the ground up — tokenization, embeddings, attention, and beyond. No lectures, no signup, just hands-on exploration.
Start the course →Courses
Zero to GPT
Build up an intuitive, from-scratch understanding of how large language models work, from derivatives and backpropagation through tokenization, embeddings, self-attention, and all the way to a working (if tiny) GPT.
View syllabus →Neural Networks from Scratch
Take the tiny autograd engine from Zero to GPT and build a real, trainable multi-layer perceptron with it — initialization, activations, backprop, gradient checking, every major loss function, batching, and generalization, ending with a 3-class spiral capstone that puts every idea to work at once.
View syllabus →AI Security
How prompt injection and jailbreaks actually work, taught defensively with real model transcripts: direct and indirect injection, roleplay/encoding/multi-turn jailbreak techniques, and the text-level and architectural defenses that hold up against them.
View syllabus →Prompt Engineering
Zero-shot, few-shot, chain-of-thought, role framing, retrieval grounding, self-consistency, context limits, and prompt chaining, taught with real model transcripts and tied directly back to the attention and sampling mechanics from Zero to GPT.
View syllabus →Agentic AI
How autonomous, tool-using systems actually work, built directly on Prompt Engineering's context-window mechanics: turning next-token prediction into tool calls, reasoning-action loops, self-directed planning, memory across steps, and the guardrails that keep an agent loop from running away from you.
View syllabus →Playgrounds
Model Builder
Assemble a small transformer from configurable self-attention and feed-forward blocks, train it live on your own text with real gradient descent, and watch a live loss curve, live sampled text, and a live attention heatmap update as it learns.
Open →Pretrained Gallery
Generate text from a real, PyTorch-trained model entirely in your browser, no setup required, or bring back your own model trained locally with the Model Builder's exported training script.
Open →Word Math
Real vector arithmetic on a real pretrained model's word embeddings, king minus man plus woman, searched against its entire real ~21,700-word vocabulary for whatever word actually lands closest.
Open →Tokenizer Playground
Compare real tokenizers side by side on your own text: character-level, BPE trained live from scratch, and WordPiece running against a real pretrained model's actual ~30,500-token vocabulary.
Open →