given a multi-purpose AI today is in the order of petabytes...
---
take those fancy 16 terabyte hard drives and and you will need 64 of them or 16 double size (2u) double that and add half more for speed and redundancy, for a total of 80u needed of server space or about 2 rack cabinets per petabyte since each 2 meter tall cabinet is 42u
---
...and you get the idea.
Kajillion is a (i kid you not: real) base 25 number that in base 10 means: 3.814.697.265.625 (or 25⁹)
This means 14.3 kajillabytes just means 50 terabytes.
But a save state of a standard AI to a standard template is at best a few gigabytes. Because all you need is the difference state from the standard template
Hence me saying that is needy.
Simone Spinozzi
2021-12-01 09:25:07 +0000 UTC
Remember when a 1.4mb floppy was big enough to save a bunch of files? Now you can't even find a thumb drive smaller than 2gb. Memory is small & cheap and software has bloated to fill the space. In the future I assume files will be 100x the size they are now. An A.I. would be an exceptionally large chunk of data, in a world where all data chunks are huge to begin with.