They are not training a whole model in a matter of days
They where working on the problem for a year using codex.
pretraining is months but they can totally fine tune in a few days
They where working on the problem for a year using codex.