Being a pain in the ass to (or in) AI companies is a real job now | Rob Wiblin | 80,000 Hours

80,000 Hours — the organisation that tells clever graduates which careers do the most good — has released an episode titled 'Being a pain in the ass to (or in) AI companies is a real job now'. Excellent news for anyone who has spent the past two years being a pain in the ass to AI companies for free, in the unpaid internship sense. The supporting evidence, per host Rob Wiblin, is that OpenAI recently fired three safety researchers: Tomek Korbak, Mikita Balesni and Jasmine Wang. All three, he says, worked on one of the company's stated top safety and alignment problems. Hold that word 'stated'.

'Stated' is doing the work of an entire communications department here, because a sentence only needs the adjective if the reality is in question. Firing the people who work on your headline safety problem while still calling it a headline safety problem is the corporate equivalent of a restaurant advertising 'sea view' because a seagull once flew past the window. The ledger is brutally simple: three named researchers out, one podcast episode in, and 'stated' left standing in the doorway holding everyone's coats. The company keeps the priority on the website. The researchers keep the anecdote.

Then there is the actual jobs board: the UK AI Security Institute, METR, Redwood Research, LawZero and Resolution are all named as places where this work gets done. So accountability now has a career ladder, a reading list, and — as we noted not long ago — a set of transparency-and-control requirements drafted by people who will not be the ones tested against them. Fine. Being a pain in the ass is honest work, and the entry requirement is mainly a functioning memory for what these companies said last quarter. The genuinely funny part is the recruitment pitch it implies: come for the mission, stay for the org chart that keeps thinning out the people who ask the awkward question.