The Future of Everything Is Lies, I Guess: New Jobs

New jobs at the human–ML boundary

A recent essay on aphyr.com sketches a future where deploying machine learning broadly creates whole new categories of work — mostly at the seam where humans meet models. The post names roles with a flair: incanters who craft prompts, process engineers who build quality‑control workflows, and statistical engineers who try to model and tame the chaotic variability of large language models. It has been reported that a surprising number of people are already employed as "model trainers," feeding expert human judgment into automated systems.

Incanters, the essay argues, are essentially prompt specialists: people who know the quirks, superstitions, and hacks that coax better outputs from LLMs. Process engineers, by contrast, focus on making workflows safe — inserting deliberate, catchable errors to force human review, wiring provenance into documents, and integrating tooling so AI mistakes don’t become legal or financial disasters. Statistical engineers would do the heavy lifting of measuring where models fail: by language, domain, input order — the kind of psychometrics for software that will take real craft and domain expertise.

Trainers, meat shields, and haruspices

Aphyr also names darker, weirder roles: "meat shields" who take accountability when systems fail, and "haruspices" — modern augurs — who interpret inscrutable model behavior. It has been reported that Almira Osmanovic Thunström demonstrated how a few fake articles can cause major models to invent an imaginary disease, a reminder that training data and trust are brittle. That emotional moment — the revelation that truth can be gamed so cheaply — is the essay's beating heart. It’s funny and terrifying at the same time. Who wants to be the scapegoat when an AI lies in court?

Why care? Because this is not just clever wordplay. The roles map to real problems companies and regulators are already bumping into: hallucinations, provenance, multilingual failures, and brittle integration. The post reads like a call to arms — or at least to realism — that the tech industry will need processes, measurement, and people dedicated to wrestling models into behaving well. Will organizations hire incanters and haruspices, or will the cost of errors force different choices? Either way, expect the job listings to get weirder.

Sources: aphyr.com, Hacker News