News for hackers. Stuff that matters.
Re:Giving GPT-3 a Turing Test by nxpnsv on Sunday July 19, 2020 @07:31AM (#23888616)
Well, yes - but humans learn from more than just reading. Kids know an awful lot of stuff even before they can read a single word...
Re:Giving GPT-3 a Turing Test by MAXPOOL on Sunday July 19, 2020 @07:54AM (#23888700)
Even in the terms of total human sensory input data rate, humans learn from a very tiny data set.

16 year old human has experienced only about 400 million wakeful seconds.

Re:Giving GPT-3 a Turing Test by so_tired on Sunday July 19, 2020 @07:58AM (#23888727)
See my other comment

> Humans get a hand crafted curriculum inputs, evolved over 1000s of iterations, in a near-optimal language encoding.

Re:Giving GPT-3 a Turing Test by DavidSJ on Sunday July 19, 2020 @09:03AM (#23889050)
At what bitrate though? At 1 MB/s that’s 400 TB, which is nothing to sneeze at.
Re:Giving GPT-3 a Turing Test by dane-pgp on Sunday July 19, 2020 @09:30AM (#23889212)
To give a sense of how accurate that estimate is, consider this reference:

"In other words, the human body sends 11 million bits per second to the brain for processing, yet the conscious mind seems to be able to process only 50 bits per second."

https://www.britannica.com/science/information-theory/Physio...

Re:Giving GPT-3 a Turing Test by akiselev on Sunday July 19, 2020 @01:00PM (#23890633)
That doesn't sound accurate. The human eye's resolution is in the tens if not hundreds of megapixels [1] with high sensitivity (meaning the compression isn't very lossy so the bits of information remain). Even if you assume only a few frames per second that's far more than 11 megabit/s.

[1] https://clarkvision.com/articles/eye-resolution.html

Re:Giving GPT-3 a Turing Test by sudosysgen on Sunday July 19, 2020 @01:30PM (#23890866)
Human vision is more complicated than that, there is some "compression" happening at the eye. But yes, it's almost certainly more than 11 megabits, more likely around a few dozen just for the eyes: https://link.springer.com/chapter/10.1007/978-3-642-04954-5_...
Re:Giving GPT-3 a Turing Test by pmoriarty on Sunday July 19, 2020 @12:06PM (#23890262)
I am very suspicious of the numbers and statistics quoted in this article, as there are no references cited for their claims, and no indication as to how they arrived at them.
Re:Giving GPT-3 a Turing Test by DavidSJ on Sunday July 19, 2020 @09:43AM (#23889305)
So quite accurate then, since 11Mb/s = 1.375MB/s.

(The conscious mind, of course, is not where most of our learning from experience happens.)