33TB of text data for a 1T-parameter model

Re:33TB of text data for a 1T-parameter model by usgroup on Sunday April 16, 2023 @06:39AM (#35588825)

Blind people learn just fine too.

Some deaf/blind people can read braille — they learn fine too.

Much less data than you might think.

Bare in mind most people can run the brain on 2500kCal/day.

And so on.

[ Parent | on HN ]

Re:33TB of text data for a 1T-parameter model by SergeAx on Sunday April 16, 2023 @08:59AM (#35589664)

> Bare in mind most people can run the brain on 2500kCal/day

Which is, in turn, 2900 W·h, or 2.5 times less than one A100 card working round the clock (300W·24h)

[ Parent | on HN ]

Re:33TB of text data for a 1T-parameter model by kristianp on Sunday April 16, 2023 @04:41PM (#35594077)

A quick google says that the brain uses about 15W of power. Or 20% of total usage of the person.

[ Parent | on HN ]

Re:33TB of text data for a 1T-parameter model by SergeAx on Monday April 17, 2023 @07:54AM (#35599439)

This is not entirely correct: brain needs oxygen, cooling and toxins removal, which are provided by circulatory, respiratory and secretion systems.

[ Parent | on HN ]