impractical.in

Explainers

Interactive, visual explanations of things that are interesting enough to take apart.

What actually runs an ML model?

Model files, runtimes, tensor kernels, CPUs, GPUs, memory bandwidth, quantization, and serving.

AI & hardware

How an LLM produces a sentence

Tokens, embeddings, attention, transformer blocks, sampling, and context.

AI & language

How a neural network learns

Weighted sums, activations, loss, backpropagation, gradients, and repeated updates.

AI & learning

How video formats actually work

Frames, chroma, motion prediction, bitrate, codecs from H.264 to AV1, and containers.

Media

How image formats actually work

Raw pixels, JPEG, PNG, WebP, AVIF, color depth, alpha, and visual compromises.

Media

The unreasonable space of a deck of cards

52 factorial orders, riffle shuffles, and why deck order is not solitaire win probability.

Mathematics

How many tiny images can exist?

The combinatorial explosion inside even a 4×4 black-and-white square.

Mathematics