Ok, outdoors, the party is just getting started... It's going to be
FUN.... Anyway, the subject at hand is, as always, AGI...
Anyway, we're so close to AGI now that it's giving me goosebumps. I am
probably wrong but I think I see enough lose parts out there that I
think that if I assemble them in a clever but counter-intuitive way I
can get over the big hump to having an AGI, albeit a very sub-optimal,
poorly performing one.
I am going to have to break your conceptual frameworks to do it and do a
buch of other things that will seem wrong/insane but will probably
actually work...
There is no point in arguing about AGI these days because the resolution
to all such arguments is to type it into a compiler, run it, and give it
an IQ test.
So, for me, it becomes a logistical problem. I need to determine how
much hardware and software I need to do the tests I need to do and then
figure out how to get it all together and working.
It would be nice to work at an AI research organization of some sort but
it turns out I'm an outcast from society. Nobody likes me. So I get to
spend the inheritance from my father's estate on hardware and bootstrap
that to total world domination and basically do nothing but be smug
about having AGI and the people who didn't hire me not... Smug is better
that bitter, right?
Digression: It turns out that Clown World (Ref Honkler the clown) was
the direct and inevitable outcome of the ZIRP from the centeral banks
around the world... Thank god, Clown World will be coming to an end Real
Soon Now....
My point is that it is frustrating as hell when all I can do is sit and
hope that someone else will eventually reach the insights I reached
years ago with no real way to do anything on my own and that I'll
eventually get access to it. My computers are beefy enough but getting
the machine learning frameworks working requires precisely matched
compiler and library versions, it's just not possible on a general
purpose desktop. =\
I'm going to need features from all of the frameworks that have been in
the news recently. I need it all to run at least fast enough to function
at least 20% real-time (or 1/5th speed). I think I can cure it of the
autism you see in the current models but I have no idea where I'll land
on the IQ scale. At this point the only goal is to successfully land on
the IQ scale, the rest is just scaling, optimization, and mass
production... Obviously I will eventually need a much closer to
mathematically optimal version for production work but at this time, I
think a hack job stands a non-zero chance of working.
https://shop.lambdalabs.com/gpu-workstations/vector
/customize?_gl=1*18ucxhp*_ga*MTQxMTkyNjY1MS4xNjc4ODkyMTgx*_ga_43EZT1FM6Q*MTY3ODg5MjE4MC4xLjAuMTY3ODg5MjE4MC42MC4wLjA.
I think that workstation will be enough to run some simpler models but the
bigger ones will
require $10+ million + datacenter + megawatts of power... =\
I mean there will be a fair amount to do on the coding side to get everything
wired up, or
I think that workstation will be enough to run some simpler models but the
bigger ones will
require $10+ million + datacenter + megawatts of power... =\
I mean there will be a fair amount to do on the coding side to get everything
wired up, or
hacked to be wired up in ways they were never intended to but needed to solve
the problem at
hand...
The tests I am targeting are being able to watch a movie, identify and describe
the charactors,
explain the plot, etc... I also want it to be able to play any video game at
atleast the level
of a noob.
--
Beware of Zombies. =O
#EggCrisis #BlackWinter
White is the new Kulak.
Powers are not rights.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T3cd927c5967dfb26-Mad2afe249d14f034096f801f
Delivery options: https://agi.topicbox.com/groups/agi/subscription