Visual AI Explainer

Every concept, animated and interactive — from a single neuron to how frontier LLMs serve tokens.

The journey: each stop builds on the previous one. Click a node to jump to its page.

1 · foundations

Perceptron

The original artificial neuron (1958). Drag data points and watch it learn a decision boundary in real time.

 2 · foundations

Activation Functions

Sigmoid, ReLU, GELU, SiLU… why networks need non-linearity, with live plots you can probe.

 3 · networks

Multi-Layer Perceptron

Stack neurons into layers. Train a real network in your browser on 2D datasets and watch the boundary bend.

 4 · transformers

Self-Attention

Queries, Keys, Values — the mechanism that lets every token look at every other token. Fully animated.

 5 · transformers

Transformer Block

FFN, LayerNorm, residual connections, multi-head attention — click each part of the architecture to explore it.

 6 · frontier

LLM Inference

Prefill vs decode, the KV cache, sampling — how a trained model actually generates text, step by animated step.

 7 · frontier

Recurrent Depth

Looped transformers: applying the same block repeatedly to spend more compute per token, with the published Huginn results and their limits.

 8 · frontier

Engram: Conditional Memory

DeepSeek's hashed N-gram lookup table beside MoE. Trace a sentence through the hash, see the gate, and watch the prefetch hide behind early layers.

 reference

Glossary

Searchable index of AI terminology, each term linked to the page where you can see it in action.