Data centres, GPUs, coding agents and the models everyone is arguing about — written for people who run operations, not for people who train models.
FEATURED INFRASTRUCTURE8 MIN
Power in, heat out, silicon kept busy, traffic moved. Five live animations of a data centre — and why all four problems end up as one number: how long your answer takes.
Physics, then biology, then the lab behind Claude — and the one uncomfortable idea the whole thing hangs off.
Co-founder, critic, competitor. The timeline from OpenAI to xAI explains more than the commentary does.
What changed this month, and what a 500,000-token window really is — the size of the desk, not the size of the library.
Why the chips built to draw video games ended up running AI — thousands of small cores instead of a few fast ones. No equations.
Read, plan, edit, test — and round again when the tests fail. The loop is the trick, not the writing.
If there is a bit of AI you keep being told matters and nobody has explained properly, say so and it will turn up here.
Suggest a topic