Writing.

4-bit quantization really is nearly free — but at 3 bits both long-context retrieval and multi-step arithmetic break, and retrieval takes 3.2× the standardized damage. Eight components get better when you quantize them. And the standard way of ranking which layers matter turns out to fail four separate statistical tests. A plain-English walkthrough of the paper, with every figure and number intact.

How a thirty-gigabyte AI model gets rounded down to fit on a laptop — scale and zero-point, calibration, granularity, integer-only inference, QAT, and the outlier problem that forced large language models to invent new tricks.

A walkthrough of what actually happens inside your iPhone when you add a card to Wallet and when you pay — the Secure Element, the Secure Enclave, tokenization, cryptograms, and the minimal code you need to write.

Public-key encryption explained through Romeo, Juliet, and a little Harry Potter — then the actual RSA math: totients, congruent mods, key generation, and why quantum computers are coming for all of it.