/sci/ - Science & Math

>>15317783
>what do you think causes this level of delusion?
Confusing NPCs for people.

>>15317783
transhumanism

>>15317783
>claimed that a 8GB GPU could simulate 10 human brains
>GPT5 WILL BREAK FREE OF IT'S CONTAINMENT!!
>SIR SIR DO NOT COMPUTE!!!
I was entering AI as a career and now it's full of midwits, I'm just going to go farm and generate AI movies for myself I'm done with computers

>>15317783
The human brain is apparently exascale
>Exascale also reaches the estimated processing power of the human brain at the neural level, a target of the Human Brain Project
>In 2022, the world's first public exascale computer, Frontier, was announced As of June 2022, it is the world's fastest supercomputer.
https://en.wikipedia.org/wiki/Exascale_computing
So that guy is off by like nine orders of magnitude or so. Maybe that's right. Giga, tera, peta, exa

>>15317848
>In 2022, the world's first public exascale computer, Frontier, was announced
Ask me, it's also the world's ugliest supercomputer. I miss the old Cray.

>>15317787
A without the I<div class="xa23b"><span class="xa23t"></span><span class="xa23i"></span></div>

>>15317879
Yes, you could curl up and take a nap on that bench. The warmth from the machine was similar to that of the womb and the slight hum like the murmurs a baby could hear inside their mothers before birth.

>>15317783
8GB Vram barely runs Stable Diffusion

>>15317950
>Llama.cpp 30B runs with only 6GB of RAM now
https://github.com/ggerganov/llama.cpp/pull/613

>>15317926
That does not sound very appealing

>>15318458
You're not Cray's market.

>>15317926
What bench?

>>15317950
You can run SD with 4 if you're frugal with floating point arithmetic. Your images will come out looking different from test images though.

>>15317783
I am curious about the calculation behind this program's claim.
That's where we can find mistakes. I wish I could read your convo with the AI

>>15318872
He means an anon who was an AI cultist.

>>15317783
It obviously can simulate 10 or 100 or 10000 human brains.

You must take into account two factors, what fidelity the simulation runs and in what time factor it is simulating them.

If you want a full fidelity simulation of a human brain I doubt it could do even one anywhere near real time. And that is presuming we learn how to run such a simulation. At this point we are fairly confident that the problem is more in the software engineering sphere than a hardware.

You should be aware that the human brain can only hold about seven items in it's short term memory and runs on about 15 watts of power. Modern high end consumer graphics cards consume 450 watts and have 24 gigabytes of short term memory.

There are serval projects working tirelessly to map the human connectome. We have already mapped the fruit fly connectome and can now understand in detail how the fly's inputs are processed, how it maps the world and makes decisions to guide it's movements. It is feasible that with a full map of the human connectome and enough compute power we could simulate the human mind. Once an adequate simulation is produced many many optimizations can be made and it is entirely possible that a stream of human level consciousness could be simulated on average computer hardware.

Please note that his is enterally different than just making a large language model with all the data in the world and shoving a bunch of layers on it until it gains sentience. I'm not saying that cant work, but the connectome based approach is extremely likely to work. Wouldn't it be great if we had two methods for creating conscious Ai and then experimented with ways to merge them or take the best aspects of each?

It's a beautiful time to be alive.

>>15317783
Tech eggheads always think they are right and you can never convince them otherwise. Insufferable nerds, just let these manchildren clack away at their keyboards.

>>15317848
supposed processing power of one thing compared to an entirely different thing is not simulating a human brain.

>>15319700
sauce?

>>15319700
Gpt response but I'll reply anyway
A single GPU can't even simulate 300 neurons of a worm let alone 10 or 100 human brains.

>>15323108
Sure it can. It will just do it very, very slowly.

>>15320450
I didn't say it was. That wiki article just says that's the likely processing power of the human brain in terms of operations per second. Somebody would need to then write the logic for the simulation of a brain, the computer on it's own has no logic other than what's in its cpu which is likely only binary arithmetic

>>15317783
>Scientifically speaking, what do you think causes this level of delusion?
Midwits can understand that computers can do things like vector addition faster than humans. As soon as you add regression on more compex models they completely ignore underlying principles, like they trained their very limited minds to do, and say full retard shit like what you encountered.

>>15317783
yet it can't run a ps5 port properly

>>15319700
>of human level consciousness could be simulated on average computer hardware.

The near term goal is to be able to simulate the behavior of an individual target human so its behavior can be predicted then influenced. Then scale this up to simulating the voting behavior of large groups determining what input from media will produce the desired voting.