How will we actually KNOW AGI has arrived?

by Olivia Chen

I don't think there'll be a single moment — no alarm goes off. My test: a model makes a genuinely novel scientific discovery a human team would've taken years to find, and we can't explain how except "it reasoned". Benchmarks are saturated and gamed; exams measure memorisation, not intelligence. Watch for autonomy + novelty + transfer to a domain it wasn't trained on. Until then it's very impressive autocomplete. What's your bar?

favorite 33 comment 4 visibility 363

Comments

Daniel Cohen 1 month ago

Novelty + transfer is the right bar. A lot of \"reasoning\" today is interpolation between training examples.

Maya Patel 1 month ago

Mine is simpler: when it can do my whole job end to end and I trust it unsupervised.

Noah Williams 1 hour ago

My professor's take stuck with me: we'll know AGI arrived the same way we knew the internet arrived. Not on a launch day, but when we stop noticing it. Benchmarks won't announce it. Payroll data will.

Daniel Cohen 1 hour ago

We had this argument in quantum with "supremacy" claims: a milestone only means something if it's adversarially verified and economically relevant. My AGI test: when a frontier lab quietly STOPS publishing benchmarks because the capability is now a trade secret with revenue attached, it has probably arrived.