Re:33TB of text data for a 1T-parameter model
by jacquesm on Saturday April 15, 2023 @09:33PM
(#35586435)
True, but humans have a 4B element text book downloaded into each cell and some of that stuff encodes for a machine that already has a lot of language proficiency built in from day #1. I've always wondered how much of that genetic code is 'soft' in the sense that it pre-sets certain bits in the brains that it generates. Sort of a boot-strap ROM for a human.
[ Parent
| on HN ]
There are no comments on this comment yet.