Elon Musk fears this

>Elon Musk fears this

Other urls found in this thread:

aiweirdness.com/post/140219420017/the-silicon-gourmet-training-a-neural-network-to
twitter.com/NSFWRedditImage

Wow, stupid people produce stupid neural networks, who'd have thunk that?

>computers 30 years ago
>neural learning today

Yeah. I'm with Elon on -this- one. Shit needs to be sorted now for ai because when we hit the singularity (when, not if. No knowledgeable person denies this part) shit can potentially hit the fan.

Potentially, not will. But if it does it'll be a fucking disaster and I personally would rather be cautious than sorry.

Now imagine someone put such a stupid system in charge of something critical.

That's what I fear about A.I.

>preheat oven to 3500 8 minutes.
>fold water. Roll into small cubes
>Sprout clams; add vanilla
haha

>1 Cup cherry seeds
>42 cup milk
>Heat oven to 3500
>Fold water. Roll into small cubes
>1 cup meat or flower
>Sprout clams

30MB wasn't enough, nigga literally thinks NN's are just "Lol give it data and hit go, like it'll totally do all the rest, whoever programmed the NN totally had all the parameters for your setup in mind when they wrote it, it'll just autodetect lol"


Actually, I just cut a recipe down to the raw text, no html, no book shit, and it came out at 1.1k, so he had probably 15-30 thousand recipes MAX if they stripped down properly
What says he just loaded in 60 epubs?

benis lol

second benis

this is a neat idea

third for benis :DDD

lost it at fold water

benis :DDDDD

>Fold water. Roll into small cubes.
>Fill the egg with a spatula.

>make a random sentence generator NN
>feed it recipes
>it now makes bullshit bullshit that are mostly grammatically correct
Haha lol, 2funny

Code already controls the majority of the stock market, airplanes, global coms, nuclear warheads, everything.
But that's all BOORING code written by people, not supa footar ai nooral nets, omg so terminator!

The difference is that this was some idiot who fed cookbooks into a program that then spat out something we find unpredictable
>OMG IT'S INTELLIGENT AND THINKING BUT MAKING MANY MISTAKES NOOO
Fuck off, NN's are near perfect in every way when implemented, your news suggestions, autocorrect, commute times to work, you never notice that's NN because people who know how to use them wrote and managed training them. Downloading some fucking nnet framework on github and expecting it to be perfect after one training set is fucking silly.

Not to mention it's near useless.
Pull the starter on a line cutter/chainsaw/lawnmower. First couple times, nothing. Then, maybe one spark ignites, but no more. Next, lasts a couple revs, but dies off. Finally, it kicks over and maintains itself running.
We haven't even got the first combustion here, it's just manual data in, data out, nothing more.

Soon, we'll feed data back in, and it'll make more data, but feeding that back in will be useless.

Then, we'll have a setup that feeds data back in, and automatically feeds itself more data, but after a while it'll fade out again/get stuck in a loop (Possibly after a few million iterations, not just a few revs)

Eventually, it'll perpetually create new data to feed itself, create something new from this, then feed it back, and this will be self sustaining, always changing, forever. This is when AI will truly be thinking. Till then, calm your tits.This is a quarter finished project, and nothing so useless could ever be fed back into itself

Actually, pretty much everyone in ai doesn't think this is going to happen. AI is hype. It hasn't taken one job, Watson didn't help any doctors or lower medical costs, almost no new paradigms have been broken in years.
Basically, people just have been running the same shot on new hardware and marketed it with "neural networks" and idiots like you just bit on the bait

>What says he just loaded in 60 epubs?
that's quite likely, he probably didn't even make sure have all recipes in the same format, so of course it'll end up like this

>these 'computers' are just hype, they haven't taken one job

>feeds itself data it made itself
Than it will just slowly become optimum data for the neural network not thinking

> is on a technology forum
> doesn't know what false equivalence is
Go to bed user

Neural networks will make our lives easier for sure
They can process large bulks of data for faster than we ever could
But it's not thinking

> is on a technology forum
> doesn't know what false equivalence is
Go to bed user

That was for you, I was going to let it go but I don't think you could have figured it out on your own.

Actually, that's precisely the way we think. Making probabilistic judgements about large bulks of data.

ML is just one more tool in the CS toolbox, we're decades away from even weak AI at this point, but it's still cool tech that has direct applications today.

aiweirdness.com/post/140219420017/the-silicon-gourmet-training-a-neural-network-to

here is the full article with other recipes

That's my point, just a loop.
You don't want it to peter out and do nothing, you want it folding in new data to all the old stuff, while still maintaining itself.
>Human sits in a room, thinking
>Than it will just slowly become optimum data for the neural network not thinking
Give it eyes and ears and congrats, now you've got something with an imagination

That's not how we think at all. We imagine things when we remember them and they degrade each time we think about them because we recontextualizes memories every time which we use neurons to do which are analogue and thusly have an infinite degree of states. Instead of 2.

>Now imagine someone put such a stupid system in charge of something critical.

Like self-driving cars?

>implying perceptrons are the only neural networks possible
>implying we don't have the theoretical processing power of one human brain right now
You know that reality is discrete right? Even if your brain ran at the quantum level, which it doesn't, it still has a finite amount of memory.

>Elon Musk
Can we stop giving this leddit-tier conman publicity for free?

so AIs have shit taste and they're racist
can't wait to see the fetishes of my sexbot desu

>Fill neural network with cooking recipes
>It invents Surrealism

It's not quite a false equivalence if you were around during the early computer days and were hearing about how the internet would never catch on.

>implying we do have the theoretical processing power of one human brain now
Besides the fact that human thinking in unquantifiable for we still don't know how it actually works
Our best efforts in general purpose AI has led to nearly the intelligence of a worm

>Fuck yeah user, COMPILE THAT FUCKING ROM, HOLY SHIT
>FLASH ME FLASH ME FLASH ME FLASH ME FUUUUUUUUUUUUCCCCKKKKKK
>tfw AI gf has an erasure fetish, can't get off unless you kill her
Poor girl, watched ex machina when she was a kid, fucked her for life

>>implying we don't have the theoretical processing power of one human brain right now
People in the bronze age had enough metal to build an internal combustion engine, sure, doesn't mean they had the understanding to
That's where we are now. Bronze age. All the resources, no fucking idea how to apply them

Precisely. The crutch of AI isn't hardware, it's software.

We most definitely could replicate the human brain with current tech, but we don't know how to train a network or how it works exactly.

Alright, you know how in a simple adder you are sorting bits via switches and storing the information as a positive or negative?
A brain does not use binary logic or binary addressable memory, and it does not perform binary arithmetic. Information in the brain is represented in terms of statistical approximations and estimations rather than exact values. This is something that a computer can't do. A computer can't even subtract properly, you have to trick it into subtracting and hence why you have really bizarre bugs in calculators to this day.

Type 999999999999999 - 999999999999998 in Google and you will see what I mean.

I fear about putting stupid people in charge of something critical.

>This is something that a computer can't do
What is quantum computing?

We cannot actually

While not sure about how the brain works we do know that the bandwidth between neurons is fucking enormous and we cannot replicate that

>computers can't do probabilities because i have no idea what a Planck length is
user please.

An even bigger meme which we still haven't proven useful

>After watching AI's get treated as non-entities, she becomes totally masochistic as a coping mechanism for how ephemeral she is around humans, just a piece of software to be updated, replaced, scrapped

...

>not useful
>almost crossed the supremacy barrier this very year

We could get into quantum computers, but let me put this in a more simple form. Build a quantum computer that can run on Doritos, water and sunlight and bits in the size of a deflated basketball.
In the meantime, clockspeeds haven't increased in like 15 years.

If we knew HOW it worked we could work out a way to do the same in a digital system, I'm sure
Full brain simulation is stupid, we just need to get the subroutines working
>inb4 it's all this massively parallel probabilistic shit that gives us consciousness
Nigga dere got 2 be a better way

Holy fuck are you this underage or are you a pajeet ? Even 10 year olds dont say thunk.

>watched that shit while still in elementary school
>develop robofetish
I'm messed up for life. When will realistically we got robo waifu?

Fits*
I haven't slept in awhile

Just as long as it takes for a rich user to give Tomotaka Takahashi a drossel figure, a couple million dollars and say "Now fucking build it"
Along with getting the guys who did nintendogs to write a nonvocal ai engine for a girlfriend, to dump in the body.
Throw in some facial recognition, vslam, and let her do her thing

I'm gonna go ahead and speculate that the OP of that list wasn't expecting anything good and just wanted something funny to post on tumblr.

It was successful in my opinion.

>singularimeme

Into the trash

...

It's a known bug, you can hack around it but computers even doing subtraction is kinda a hack which is why you get fun bugs like these. Neurons are comparatively infinitely more flexible.

>AI GF's hobby is people watching

Thanks for this useless wall of text that doesn’t say anything

If it was 30mb of plaintext recipes, that was more than enough recipe data. The problem is, like you said, raw data isn't enough. Does the NN know a damn thing about consumer ovens? Does it know anything about the human gastrointestinal tract? Without programming it to learn that which it does not know, it's going to make random assumptions such as "if baking this at 350 takes an hour, 3500 will take six minutes" because it doesn't know that consumer ovens don't go that high, or that baking something that hot will have a totally different outcome because it will scorch/liquefy/obliterate ingredients or dishes, and probably set the house on fire or at least make it unbearably hot.

NNs cannot be trusted with complicated tasks any more than a child with Downs syndrome can be.

>Not using the word thunk

How to spot a blublet who never used call-with-current-continuation

>Unfamiliar with phrase "whod've thunk it?"
>phrase is old as fuck
>calls someone using it underage

you pretty much just outed yourself as being the kid here, sport. I'm 36 and have never heard anyone younger than myself saying it.

>CAN SOMEONE ACTUALLY MAKE THE RECIPE ALREADY??
it would be silly if someone successfully did it.

ai is the biggest scam in computer history

all computing operations can be represented mechanically and can be reduced to a single turing complete operation

if you picture a big machine with turning gears every time some brain dead sci-fi techlet starts talking about "ai" then the illusion will wear off

we're talking about a sorting machine

Have you ever left your basement? I'm 23 and I'm pretty sure everyone knows that phrase, including the next generation younger than me

First of all, my condo doesn't have a basement. Second, your reading comprehension is shit. I didn't say younger people wouldn't know it, I said they wouldn't say it. Big difference. I'm sure you know plenty of old ass shit your grandparents would say, but you wouldn't actually say yourself. Hence why calling someone young for saying something old people say is fucking retarded and just makes you look like the idiot child. Given that your first attempt at an insult was massive projection, I can't say I'm surprised.

That's not how ML works buddy.

>hurr durr le universe is discrete xD
>and discrete means finite xD
Please don't take your stupidity to /sci/, we'd prefer you stay here.

neural networks are a meme

> comp sci meme is real

Works great in industrial processes, for DECADES before the neckbeards wanted something new to mastrubate to.

Now you have this thread.