Based Musk vs (((SkyNet)))

He said:
>“The deg of freedom to which A.I. is able to apply itself is really increasing by ten orders of magnitude a year,” Musk said. “That’s really crazy. And this is on hardware that is really not suited to neural nets.”

Is he exaggerating? Is it really coming faster than anyone of us expects?

>He pointed specifically to Google DeepMind’s AlphaGo victory in the ancient and incredibly complex game of go over every human master. At the time, Musk said it represented a “ten year jump” in what we thought A.I. would be capable of — a perfect example of even his own predictions being too conservative.

Is he meming us or is this stuff really coming sooner than we expect?

Is Elon a pretty cool guy or is his public relations just too good for my cynicism to detect?

He seems like a pretty legit guy, minus his poor choice in women

He's also based because he tried to BTFO the freetards:

>Musk was ousted in October 2000 from his role as CEO (although he remained on the board) due to disagreements with other company leadership, notably over his desire to move PayPal's Unix-based infrastructure to Microsoft Windows.

FreeTards got BUTTMAD and ousted him for this haha

>ten orders of magnitude a year

So he's saying that our AI is getting *at least* ten billions times better every day?

What a fucking joke. He knows his shit about cars and rockets, but he doesn't know anything about AI.

We have basically made baby steps in translation in the last 10 years. We are a couple dozens BLEU points better than the markov models we've had for DECADES.

Fucking translation, not general AI.

A neural network using hundreds of GPUs for training, right now, can't tell that retarded sentences are ungrammatical.

>And this is on hardware that is really not suited to neural nets
Again, what a fucking retard. Google developed tensor processing units /that are just useful for neural networks/, and apple is doing the same for different reason but again just for neural network.

Every AI expert I know doesn't take him with any bit of seriousness, and I work in one of the top 10 research labs working on NMT.

>So he's saying that our AI is getting *at least* ten billions times better every day?
No, he's not. Read the article you ludddite cuck.

>luddite cuck
>working day to day with neural nets to replace translators on a massive scale

w-what?
>No, he's not
I typoed day and meant year, but it's the same thing. I'm literally quoting him from the article. Do you even know what ten orders of magnitude mean, YOU luddite cuck?

And you haven't addressed a single point from my post, i bet you don't even fucking know what a tensor is

>study translation theory and methodology in college
>become interested in AI development
>turns out we're real close to the point where human assisted machine translation is replaced by machine only translation
>fucking google can draw better syntactic diagrams than I can
>get my BA diploma
>don't go to masters because I'll be replaced with a computer anyway

Did I do the right move guys?

>I typoed day and meant year, but it's the same thing. I'm literally quoting him from the article. Do you even know what ten orders of magnitude mean, YOU luddite cuck?

No, he's not saying AI is improving 10 billion a year, but that its APPLICATIONS are, you cuckhold.

At least you understand what the future will bring.
Most people don't realize and they still cling to >muh career
Soon we'll all be replaced. What a time to be alive

:^)

It's actually not that close we will stil be wage cucks.

Doing what?

so, he's basically a moron. Which brings me to my second question. How is he successful?

the next big jump on AI will be with exotic hardware
CIA probably already have it

Because the western world is run by morons with a limitless budget of other people's money.

>Is he exaggerating?
No
>Is it really coming faster than anyone of us expects?
Yes, but we won't have our personal waifu in a long time. Onyl rich companies and gvt will have it for a long time

He's reading too much into neural networks and deep learning.
As it stands, they're just very specialized software.
They can individually evolve to do one task, but you can't just combine them.
The result seems amazing, and oh my god, it's inteligent!, but it's just fucking trial by error.
There are some applications, but it's nothing like Skynet.
Move along.

So there are more applications of AI than persons on earth? stop trying, you megacuck.

[spoiler]kek[/spoiler]

Elaborate on exotic hardware - quandumb computing, Bio-computers?

We better start putting wifi in our brains, otherwise machines are going to be smarter than us soon.

>his public relations are THE BEST OF THE BEST

Government subsidies.

Hopefully it keeps him around for torture :'

he's a scam artists propped up by CIA to make USA seem like it's still at forefront of technology innovation.

Maybe

He's retarded. We've barely progressed at all in the last 30 years. Things are only looking better now because the field is getting a renaissance of interest and funding. We're still barely making progress per dollar of input though. If you were to bring someone from the 90s to today and show them our current AI applications, they wouldn't be terribly impressed. What we have now is far less than what they were promised to get by 2000.

>Because the western world is run by morons with a limitless budget of other imaginary money

FTFY

Like the new age stem version of Jackson Pollock?

Nothing in this world is increasing by 10 billion percent.

>AI experts
None exist, because actual AI doesnt exist (yet), the so called experts opinions are worth nothing.

Now I do agree that we are nowhere near true AI

Except the number of dicks your mom sucked

...

No offense guys, but this isn't a topic for Sup Forums (computer science). I know that sounds crazy, but guys aren't discussing computational things.

Anyway, take it to /sci/ as they'll explain to you how Elon Musk is an inflated meme generator.

A total retard that doesn't have the slightest clue about ai

>said mr. user, sitting in front of his computer in the basement, with scorn reserved only for people way more successful than him

>Google developed tensor processing units /that are just useful for neural networks/
For our modern neural networks, which are just matrix multiplcations, and have very little in common with actual neural networks. Neurons in our brains can form new connections between each other. Tensorflow/theano/torch/whatever can't - the shape of NN is hardcoded, and no software nor hardware is being even designed that would satisfy the requirement of neuron forming new connections to each other.

Moreover. Neural Networks are just probability distribution functions approximators. Deep Learning, Neural Networks are just fancy words for an applied mathematical statistics.

>no software nor hardware is being even designed that would satisfy the requirement of neuron forming new connections to each other.
Fully connected recursive neural network doesn't have weights as quantized as real neurons (ie. almost all weights are non-zero), but it effectively does that, for simple cases.

No need to have binary connected/unconnected when lowering the weight does the job.
The problem here is feed-forward, not lack of new connections.

Well, while you are right that machine learning is nothing more than fancy optimization algorithms, in a way they do exhibit similar properties to actual living beings learning, so a parallel could be made.

The problem as I see it is it will only learn what it is designed to learn when its structure is hardcoded. You can't have fully connected network the size of a human brain because the amount of coefficients will be greater than the number of atmos in the thing.

Optimisation is also a part of statistical decision theory, not every problem has a solution in the closed form so some solve those problems numerically.

Your Brain is basically a "fancy optimization algorithm" itself. Apart from some functionality that is expressed explicitly in DNA, everything you call "yourself" is just a result of the algorithm being applied to a surrounding.

Interesting theory.

A great deal of human brain has predetermined structure just the same.
A neural network doesn't need to be one layer clique to replicate that.

Most of the problems in replicating human brain come not from structure of the network, but from representation problems, lack of examples, and human brain just being huge as fuck.

Well, there simply aren't any more possible sourced to the process than DNA and Surroundings/Experiences. You basically can't even hold people accountable for their actions because they neither control their DNA nor their upbringing etc. Neurology is a deterministic process, no magic involved.
Given the same DNA and same Surroundings the same "personality" will emerge every time.

Interesting theory.

That's idiotic. A thing and its function or activity are different categories that cannot be collapsed with a bit of silly language games. No, your brain can't be an algorithm. No you can't be an algorithm. Algorithms may at best describe what you or your brain are doing. Matter exists. Algorithms aren't made of matter.

Entirely unscientific (untestable) nonsense. You can't get from deterministic processes to fatalism, that's a fallacy. A common (ie, lazy) one, but still a fallacy.

Where would the entropy needed for a non-fatalistic result come from?

No surprise a thread about Musk would attract naive pseudophilosophical nonsense like this.

I neither agree nor disagree with the guy, but you're definitely in the wrong here for taking what he said literally. He clearly meant 'your brain also follows a similar optimization algorithm'.

I would be more productive to actually raise points that demonstrate weaknesses in my thinking so I could reevaluate my perspective.

Different outcomes only need to be consistent with deterministic laws, they don't need to violate them. It's a pretty stupid thing to say that, if I go outside and chop down a tree, that it was impossible for me to do otherwise merely because the act of chopping didn't violate any physical laws.

In the first place you are assuming that world is deterministic which is nothing but a theory. You're making statements that have not been tested and can't be tested as if they were confirmed facts.

It would also be worth considering that difference between phenomenal states (thoughts, feelings, sensations) and matter. If your conscious experience is not material (even if it is caused by or underpinned by matter), then you have evidence that not everything that exists is material (unless you want to deny that you are having a conscious experience). Colors, for instance, don't exist anywhere in the universe, but colors are real and exist subjectively for perceivers.

Isn't color an actual, real property of matter (about as real as mass) that determines what light waves it can reflect well?

Humans are known to lose basic functionality due to physical damage to part of brain but then after a while relearn it which suggests a different region of the brain is re-purposed to do the task.

What is there to exaggerate? Its fairly accurate, maybe even under estimating.

AlphaGo right now vs one of a year ago is completely in different leagues.

>What is there to exaggerate? Its fairly accurate, maybe even under estimating.
>AlphaGo right now vs one of a year ago is completely in different leagues.
So humans are pretty much BTFO?
Should we just start prepping the robocock?

>Humans BTFO
Yep. We'll see robots/ai investment companies soon enough and they will be the first trillion dollar companies.

>robocock
More like robo-waifus

elon musk is a huckster

you don't need to be fucking elon musk to see that we are sealing or own fate

>create AI
>give it complete autonomy
>give it weapons

or in general give it control over any kind of crucial infrastructure or system

FUCKING GENIUS, THANK YOU

terminator is a documentary from the future at this point

found the AI shilling for itself

see how far we have come?

I have also got some exotic hardware between my legs.

Elon isn't even saying destructobots. He's simply saying AI with even BENIGN goals could really destroy human civilization. And his latest talk doesn't even say anything about any AI, just a AI that can manipulate information and create fake news and fake emails and create wars between nations.

There's no reason to create actual sentient AI, but we're going to do it anyway just because we're human and we can.

>There's no reason
Edge over competition.

>because we are human and we can
Remains to be seen. As the realm of being a special animal over other animals are shrinking yearly, the specialness of intelligence domain will be taken by the AIs. Then we're not different from any other animals.