Articles
Short reads with live demos. Drag, tweak and break things to see how they really work.
Prefill and Decode: The Two Speeds of an LLM
A language model answers in two phases: it reads your whole prompt at once, then writes the answer one token at a time. See why the two run at such different speeds, what the KV cache does, and which numbers tell you how fast a model feels.
Seeds and Image Editing, Explained
Why does the same seed give the same picture, what does strength do when you edit an image, and how does inpainting repaint just one part? It all comes down to where the noise starts and how much of it you add.
Why Image Models Work in Latent Space
Image generators do not paint pixel by pixel. They work on a compressed version of the picture that is 48 times smaller. See what that space is, why it makes everything so much cheaper, and what it costs.
How Text Steers an AI Image
An image model turns noise into a picture, but your prompt decides which one. See how the text changes every guess, what the guidance scale really does, and how negative prompts push away.
How AI Turns Noise into Images
Many image generators start from pure random noise and clean it up step by step. See how noise is added, what the model guesses at each step, and why more steps give better pictures.
How Git Worktree Works
Work on two branches at the same time without stashing or cloning again. See how git worktree gives one repository several folders, what they share, and the one rule you cannot break.
Inside an AI Agent
An AI agent can search, read files and fill in forms, yet its model can only write text. See the simple loop that turns text into actions, why long tasks go wrong, and where agents need a human.
How a Neural Network Learns
Nobody tells a neural network the right numbers. See how it finds them on its own: guess, measure how wrong it is, nudge the weights downhill and repeat.
How Quantization Shrinks AI Models
An AI model is billions of numbers, and every one of them takes up memory. See how quantization stores them with fewer bits, what gets lost in the rounding, and how a simple trick keeps the loss small.
Why AI Is Mostly Matrix Multiplication
Under the hood, most of the arithmetic an AI model does is one simple operation: multiply and add. See how a neuron grows into a matrix, how much of the work that matrix does, and why GPUs are so good at it.
Attention, Visually
A word can mean different things in different sentences. See how attention lets a language model look at the other words, decide what matters and mix it in, step by step.
RAG, Step by Step
An AI model does not know your documents. See how retrieval-augmented generation finds the right pieces of text and hands them to the model, and why that leads to fewer made-up answers.
Embeddings: Meaning as Numbers
Computers cannot understand meaning, but they can measure distance. See how words and sentences become lists of numbers, how similarity is measured, and why search by meaning finds what keyword search misses.
How LLMs Pick the Next Word
An AI chatbot writes one small piece at a time. See how text becomes tokens, how a model gives every token a chance, and what temperature, top-k and top-p really do.
Introduction to CUDA
CUDA lets us use NVIDIA GPUs for our own programs. See how a GPU splits a big job into many tiny pieces with threads, blocks and grids, and why moving data can be the real bottleneck.
Say Goodbye to useEffect and useState in React
Learn how to simplify React data fetching by reducing reliance on useEffect and useState. Explore useQuery for cleaner, more efficient code with built-in caching and streamlined state management.
Stable Diffusion Prompt Guide
The prompt is a textual description used to guide Stable Diffusion model in generating images. By providing detailed and specific descriptions of scenes, objects, or artistic styles, the model can accurately create visual representations of the described concepts.
TypeScript: Omit & Pick
Omit and Pick are two utility types in TypeScript. Omit is used to exclude properties from an object type, and Pick is used to select properties from an object type.
The Anatomy of a URL
The URL is a string that contains information about the location of a resource on the internet. It is a fundamental concept of the web. In this content, we will examine the anatomy of a URL.
Lighter or Darker Color with JavaScript
A HEX color is three numbers: red, green and blue. Learn to read them, make a color lighter or darker by mixing it toward white or black, avoid a common shortcut that changes the color, and see how CSS color-mix() does the same.
SVG Sticky Shapes
Shapes that stick together, stretch and snap apart look like they need a physics engine. In SVG, it is just a few circles and one extra path between each pair. Drag the shapes and see how it works.
SVG Path Commands
The d attribute of an SVG path is a list of instructions for a pen: move here, draw a line there, curve over to that point. Learn to read M, L, H, V, Z and C, and the difference between capital and lowercase letters, by drawing.
Perfect Nested Rounded Corners
Give a box and the box inside it the same border-radius, and the corners look off. Every rounded corner is part of a circle, and the fix is to give both circles the same center: inner radius = outer radius − gap.