Princeton professor Zeynep Tufekci argues generative AI won't cause mass unemployment due to technical limits on non-verifiable tasks

Yes, Erdős conjecture debates are fun, but that’s a verifiable, formal domain.

Many eterministic tasks have long been automated by regular software and what remains isn’t easy prey for gen AI. (See: previous AI winters).

The LLM ability in complex language is misleading us.

The data backs your core point. Anthropic's own economists found no detectable rise in aggregate AI exposed unemployment, replicated independently by Yale and an IMF study of Denmark. But the same research is consistent on what is measurable: it's concentrated almost entirely in entry level hiring, not existing jobs. I wrote my book 16·35·65 around exactly that mechanism, not mass displacement, a generational door closing at the point of entry. The labor data is sharper than either the doom or the dismissal.

@zeynep indeed

@jackhawksburn @zeynep Yes. Aggregate unemployment is the wrong dashboard. The interesting bit is where firms redesign work deeply enough that juniors can use AI as a ramp, not a trapdoor.