Machines learning¶

In the twisted streets of Ankh-Morpork, a model goes wrong in whatever way its kind goes wrong. This is its story: the relatives it meets beforehand, the notebook it wakes up in, the road out to production, the neighbourhood that decides how it lives, the two flaws it was built on and the apparatus that grew up around them, and the uninvited guests who eventually come looking for it. Data pipelines evolve like rival guilds, cloud bills grow for capacity nobody will sign to release, and in the grander houses an update requires a sacrifice to compliance. Models learn, regress, and sometimes even survive, leaving behind a trail of technical debt, post-mortems, and slightly singed egos.

Disclaimer¶

This tale was assembled by hand and checked twice, by a person, allegedly. No model was trained on your reading of it, and any resemblance to a production incident of your own is statistically unavoidable rather than personal. Side effects may include recognising yourself in the engineer, or, worse, in the model.