Re:33TB of text data for a 1T-parameter model
by flashgordon on Saturday April 15, 2023 @07:48PM
(#35585813)
True but the machinery (is the brain) that enables humans to acquire/process/understand that knowledge is a lot more complex and went through years of environmental conditioning?
[ Parent
| on HN ]
Re:33TB of text data for a 1T-parameter model
by sgu999 on Saturday April 15, 2023 @07:57PM
(#35585870)
If we are going in that direction, I guess the few million years of evolution should also count
Re:33TB of text data for a 1T-parameter model
by flashgordon on Saturday April 15, 2023 @08:27PM
(#35586039)
No totally. Not an ml guy by long shot. My understanding is llms are foundational models which iiuc are "base" models representing all human knowledge until some point in time. Now I'd argue that you don't have to count all of evolution but just the point from when you were born. I'd say the all the sensory experiences + knowledge input from the day a brain is activated (DOB) would be in the order of PBytes if not more?
Re:33TB of text data for a 1T-parameter model
by lisper on Saturday April 15, 2023 @07:51PM
(#35585842)
Yes, but that environmental conditioning did not involve training on the contents of the Kindle library. And the results of that training are encoded in under a gigabyte.
For the record, I'm not saying LLMs are not a huge step forward. They are. But they are not -- and cannot be -- the whole answer.