I'm sure many of you are aware of this issue but I'll repeat it anyway
because I had been allowing myself to get carried away thinking about
what kind of hardware I'd throw at our resident genius' brain simulator.
The fundamental problem is this paradox:
A. We do not know what kind of hardware an AGI machine will require. It
is still completely possible that it turns out to be nothing at all like
the tensor cores you will find in a GV100.
B. It has been shown that specialized hardware can MASSIVELY accelerate
AI (which may or may not = AGI) algorithms to the point where previously
infeasible solutions become not only possible but commercially viable.
It is close to certain that any general purpose architecture, whether it
be MIPS, Alpha, MMIX, ARM, Power, or even common x64 will do a perfectly
medeocre job at doing whatever it turns out to be.
However, a specialized architecture will either be fantastic or absolute
crap to the point of not being able to run it at all.
So the critical question is what exactly does AGI hardware look like and
how much of it do we need to bridge the gap between what we can do with
general purpose architectures.
I think GPUs and FPGAs can be considered relatively safe things to shop
for. At least you can waste your life playing video games with a good
GPU if it doesn't pan out for AGI work. =P
However, anything advertised as a deep learning accelerator -> BLEH
because I think deep learning only has a few months of life left to it
as a research paradigm and then the field is either SHIFT or WINTER.
--
Clowns feed off of funny money;
Funny money comes from the FED
so NO FED -> NO CLOWNS!!!
Powers are not rights.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/Te56d98b1cab7ca45-M749e4150c1dec15767ee6b0c
Delivery options: https://agi.topicbox.com/groups/agi/subscription
