原推:Every single day https://t.co/d1rdtu7PXH
in case you were worried
about running out of language data
with which to tune
larger and larger AI models
Humanity produces a volume of spoken language
that is ~500x larger
than the corpus of internet text
that was used to train GPT-3