so it is over for AI development or is it just getting started Posted on January 1, 2023 by Anonymous so it is over for AI development or is it just getting started
the only decent replies ITT.
I wish that was me
Just getting started. Neuromorphic chips are just being developed and not used for anything yet. All we have seen so far has been done on old hardware made for WindowsXP, videogames and browse internet. At no point was AI development a concern for manufacturers, but now it finally is.
>Neuromorphic chips are just being developed and not used for anything yet.
They're attracting a lot of industrial attention, but haven't made it to consumer products yet. The interest in them relates to either being able to run models with far fewer resources, or being able to put models that can learn in resource constrained situations. For example, I know that neuromorphic systems (especially ones with runtime dynamic virtual synapses) can learn complex things from only thousands of examples, rather than needing millions of them, and can do it while they are being used. That's a huge change from the current standard machine learning shit.
We have about 1 or 2 years of cool stuff and then plateau for a long time (maybe forever)
Will it at least replace all 2D artists, 3D artists, lawyers, codemonkeys, comic book writers, article writers, musicians and other jobs people posted here about?
any person born before 2000 will laugh at getting judged by a computer, call it a loser and a selective murderer and illuminati their way out of everything like how god promises
Arent the people most mad about AI all zoomers? Mostly because AI destroys their dreams of doing art/videogames/programming/being an influencer? While millenials who are not in crippling debt are mostly ok with it because they already had their fun doing art and are old enough to be senior programmers who wont be replaced by AI?
No because like self-driving cars it's just good enough to work but just bad enough that you have to hand hold it the entire time. And the amount of work to get it from 90% good to 100% good is insane.
>We have about 1 or 2 years of cool stuff and then plateau for a long time (maybe forever)
I can't draw, compose music, or write good stories with interesting dialogs. If the AI can help me to bring my dreams to reality I don't care if it plateaus forever.
cool, we have the AI doomposters
anyone cares to object?
You're a pedo, how's that for an objection gay?
my model/prompt has no dicky in it
it's gonna be over when it learns about garden gnomes
>task performance plateaus at human-level
it's over AIsisters...
Does not matter. AI art at human level is perfect art. AI lawyer at human level is cheaper lawyer, so pisspoor fucks will prefer it in courts, AI doctor at human level is still better then shortage of doctors. Overabundance of doctors would bring the prices down for healthcare. Also, when we finally find out how to define better training, we can easily achive god-like AI skills in certain tasks. In games where the rules are hard set, AIs are performing insanely well against humans, for example the Chess AIs being so much above and beyond the best masters of chess
Why do humans want to kill themselves like this?
I am ok with it as long as it kills commies, liberals and juice
AI gays don't really want AI future. Basically nobody wants AI future, they are just shilling for it so that we get the transition fast enough. We cant just decide to not develop certain technology, we as humanity are slaves to progress, we will develop a thing the moment we can and it becomes affordable and profitable, and nobody can do anything about it, no number of philosophers or politicians can do anything about it. Automating art is fundamentally bad thing. Nobody asked for it, most people are not happy with it, artists are seething over it, and the improvements it brings are that you can make img2img memes as well as not pay people for equal or worse product. But it still happened and artists cant do anything about it. All the media companies will burn as they go bankrupt from text2video, while they themselves are financing their own doom. Trying to outlaw this tech is useless, because it will still happen. Genetic engineering on humans is illegal all over the world and pretty much does not happen, but that is only because it is easier to enforce. You need big labs, you need to raise those GMO children, you need to abduct the people and you need fuckton of scientists. AI lab only needs 14 or less people unless it is some real innovation like Google, Meta or OpenAI. AI will be invented, and it will be end of history, and nobody can do anything about it. Everyone warned about it with endless sci-fi media depicting AI to be worse then Satan, and it still will happen, so why not embrace it and try to shape it in a good way instead of bad way.
if you aren't satisfied with the current state of affairs, might as well accelerate
>when we finally find out how to define better training
We have a pretty good idea, but building out a totally different way of doing things is really hard as the tech affordable at the silicon level by researchers is generations behind.
The important thing is timing patterns, not values. That most AIs use floating point values right now is an accident of history (it's correlated to what mid 20th century instruments could measure, not what neurons really do). The amazing thing is that standard machine learning has made so much progress in the first place, given how massively wrong it is.
GOD DAMNIT WE WERE SUPPOSED TO ACHIEVE SINGULARITY
we just need the AI to make its own training data, rather than rely on human made data
That's fine. As long as you can put a billion human equivalent of thinking on some task in a reasonable amount of time, that will be a big victory.
It requires huge computing clusters
Youll rent gpu farms and youll be happy
All those task in the chart are capped at 100% accuracy, so is only natural for them to plateau right after reaching human performance. Now compare it to create a more energy efficient CPU: we are still far away from the theoretical hard limit, so there is a lot of room for AI to do their best.
That data is very promising especially language understanding?
>y axis is petaflops
What did they mean by this?
Why would it be over? The only thing that's over is regular people being able to afford the resources to train new models.
I want to fuck his bussy
Brazilian Catgirl is a fully biological female with no sexual deviations.
It's gonna continue but way fewer people are going to be talking about it a few months from now. It's like how some people thought NFTs were going to be all the rage for the foreseeable future and then interested fell sharply after a while. I think AI will have more popularity spikes down the line, but it's not gonna ride this high for that long.
Its over. I have received word from a most trusted source that the crown has heard the complaints of the people of twitter and will be dismantling the computer machines in a fortnite to avoid harming the most important members of our society, porn artists!
I find it funny how everyone keeps mentioning porn artists as the only people being affected by AI art, when they are the least affected group of artists out there aside from banana-on-wallgays. Ai still sometimes struggles with making good looking penises and natural looking vaginas. It still fucks up poses, and all the best AI models are SFW only. From all the artists I know the only ones who saw huge dip in commissions were DnD gays. I mostly blame people for it because their commissions are even more generic then coomer commissions. You get to choose any magical race doing literally anything, using whatever type of class you want, even making your own and bend the rules however you want, and yet 95% of all DnD gays just choose to play as big beard wizard, elf archer and orc barbarian. The monsters they fight? Goblins, skeletons and red dragons. You can make up literally whatever you want, ask people to draw monster from your complex in-depth commission and yet they still choose to do generic shit they can find from just google.
Coomerchads on the other hand still make fursonas and futanari monstergirls shipping as if nothing changed.
Im always laughing at that poor blurry 'AI art'. You are done for, robots, haha!
n^3 per n*n [->] 2(n/2)^3
AI cannot predict what people are going to draw in the future, due to cutoff points in their training data.
ask you aids ridden boyfriend you disgusting gay.