Anonymous B joined in and replied with this 2 years ago, 11 minutes later[^][v]#1,263,146
It doesn't need to be stored. It can be generated while streaming and the random seed could be saved to reproduce any given work.
Energy limitations are a social, not technical problem. With fusion power, there is virtually endless amounts of energy waiting to be used. And even if we can never master fusion, we already have a runaway fusion reactor literally within sight for half of every day.
Anonymous B replied with this 2 years ago, 23 minutes later, 9 hours after the original post[^][v]#1,263,209
@1,263,194 (boof)
Yes. Generative networks output a probability distribution for the next event in the sequence. For text, that'd be a distribution over the next letter or pair of letters. For images, it could be pixel values. In any case, it's up to you to decide which event to actually select. Most of the time it's done pseudo-randomly. So if you start with a pseudo-random number generator seeded with a known value, then the entire process of running the network, selecting an event, and then repeating would be entirely deterministic and repeatable.
Anonymous B double-posted this 2 years ago, 2 minutes later, 9 hours after the original post[^][v]#1,263,211
@1,263,196 (dw !p9hU6ckyqw)
No one said this had to be done overnight.
> Lol energy grids have maximum capacaties
Right, I forgot that we were talking about energy grids here! Completely immutable structures bound to be unchangeable by the laws of the universe itself.
it really doesn't. starting from zero and going all the way through to getting permission to operate takes about a decade and that's an effect of government red tape. there are companies now that plan to retrofit existing coal power plants with nuclear reactors to shorten the approval times. the "building reactors" part isn't the constraint at the moment
Anonymous B replied with this 2 years ago, 5 minutes later, 19 hours after the original post[^][v]#1,263,339
@previous (D)
Also funnier given the historical context of how quickly the first atomic and hydrogen bombs were constructed. Now decades later, somehow the idea of building a run-of-the-mill reactor seems like a colossal mega-project to some.
Anonymous D replied with this 2 years ago, 6 minutes later, 19 hours after the original post[^][v]#1,263,343
@previous (B)
and, to get back on topic, if your AGI is instructed to make 100 years of TV and it needs more energy, it might just build whatever it needs and deal with the consequences later. that's probably more efficient than wading through a decade of red tape.
Anonymous B replied with this 2 years ago, 6 minutes later, 19 hours after the original post[^][v]#1,263,345
@previous (D)
I half-seriously hope AI just figures out cold fusion for us.
It's also why I think it's pretty funny that people get worried about AI deciding to destroy humanity or some shit. It just reeks of another excuse to regulate it to death and for another classic regulatory capture by the already established players.
AI has no need to annihilate us. We will compete for exactly zero resources and AI need not be restricted to earth or our primitive means of energy extraction.
Yes, anything could happen. But Allah could also decide he hates digital waifus and kill us all. Not something I'm going to lose sleep over or support over-bearing legislation to try to prevent.
Anonymous B double-posted this 2 years ago, 4 minutes later, 19 hours after the original post[^][v]#1,263,350
Also, I have no doubt more energy efficient chips will be developed using increasingly specialised hardware. Google is already doing this with their TPUs, but it's really just the begining.
The human brain uses about 15 W. I don't know what ChatGPT uses, but it's a fuckton more. A single GPU card can use hundreds of watts and that's not going to power anything resembling full human cognition at this point.
Remember when computers used to be the size of a large room and use just as much energy? Yeah.
Anonymous D replied with this 2 years ago, 1 minute later, 19 hours after the original post[^][v]#1,263,351
@1,263,345 (B)
the doomers are just freaked out about the concept of tech singularity. the machines will be doing rad shit that we won't understand and doomers can't handle that idea. like imagine you wake up and the AI has fixed up your local power grid, and you don't know why but hey that's cool, thanks AI... but it just wanted to make more TV shows and waifus. AI doomers don't like that they won't know what's going on, even if it's something cool and objectively good
Anonymous B replied with this 2 years ago, 3 minutes later, 19 hours after the original post[^][v]#1,263,353
@previous (D)
lol. In all seriousness, Google (Deepmind) wrote a paper a little while back about how they were using AI to re-design their server farms for more efficient energy usage.