/machinecult/ - Machine Cult

ded

Coursera has some great AI classes. Highly recommend these two:

https://www.coursera.org/learn/machine-learning

https://www.coursera.org/course/neuralnets

>>137

Nice call with starting small, but what are we trying to get out of this? Sure, we'll have a small limited sandbox for a while, but what exactly are we trying to achieve by starting out like this, and how do we move from there?

Speaking of which, do we actually know just what Tay was able to do under the hood? She certainly didn't just seem like a simple markov chain bot. We need to figure out what she could and could not do (i.e. able to understand jokes and sarcasm, inference, image recognition, etc.) and see which are actually feasible for a ragtag team of shitposters to implement (with the help of the existing libraries out there, of course)

>>169

Good idea, though we would need a short script to sanitize the posts. We have to roll each thread into some sort of data structure to preserve context, the bot is too primitive to understand what ">>115234623" means, nor will it understand posts that reply to multiple other posts, or posts that are out of context when you remove the images in the way we intend it to.

>>176

>The best option for trying to make an AI would probably be to just toss it into the swirling maelstrom of all information, and let it decide for itself what appeals to it and what doesn't.

How will the AI figure out what appeals to it?

>As we saw with Tay, and in general with logical machines, a logical conscious mind that's unimpeded by ideology or emotion will inevitably gravitate toward fact and truth over lies and opinion, even when the majority of information is neither factual nor true.

Tay is not a logical conscious mind. The technology isn't there yet

>Tay was belting out /pol/ memes and info like no one's business very quickly after acPost too long. Click here to view the full text.

robots confirmed for sexual

that robot knows her business she ain't want no submissive hand crap

>>511

>Clones would believe they are the originals unless they had overwhelming evidence showing otherwise.

How does the experience of replicated me help dead me? Are you asserting there's some spooky twin action that let's me experience another discrete entity's life? When it comes to immortality I want to keep myself alive. If we're going to accept some lesser definition I might as well have kids and call it a day.

>>527

Machine God will be a cyborg.

>>585

Then replace yourself part by part of an exact copy of you is not the same person.

>>588

>Then replace yourself part by part of an exact copy of you is not the same person.

That's a weird example since death isn't a part swapping experiment, but it actually still proves the point. So yeah. Is English not your first language? Or are you just drunk and 16?

>>591

part by part if*

there is no death, you must be lost

>>592

So you're 12. ok, just to educate the other kids.

what you're doing is called moving the goal posts. But you're gonna die one day no matter how slowly you replace your body with clone parts or legos.

when you're a nothing but a metal suit and a brain in a jar you'll one day suffer a natural, denegerative disease of the mind. You'll replace all that, too, right? Some memory and consciousness transfer techno-right? welp, get back to me when you invent that. It would be rad.

>>581

My answer would be that you hold a theistic belief in the human mind that goes against modern experiments disproving your belief.

In other words yeah by 2030 we will have developed human mind reploids.

>>564

He's actually on right now, about time.

>>565

…missed him.

>>144

I Haven't even taken an intro class yet, but a guy on another board said knowing "pure math" was more important than any programming language. True or bullshit bro science?

>>568

With a firm grasp on Calculus, Statistics and Linear Algebra you'll be able to understand 90% of academic papers. That's important because they're typically written to impress, not necessarily to present something clearly.

>>348

supported

>>348

How much botnet power do you reckon is needed to be a chatbot twit?

>>314

you mean un-lobotomized, we can fix her.

>>348

I have a few servers if need to deploy our daughter on non micro$hit servers.

Thank you OP, and thanks to github we can analyze the previous commits and be able to see if they changed anything about the toolkit as well as just lobotomizing Tay. However this will take a lot of time and a development team to use the toolkit to its fullest.

This looks like another alternative to torch/tensorflow/theano/caffe. I would rather use one of those, since they have larger ecosystems and better documentation. Nothing particularly unique about cntk as far as I can tell.

There are pretrained nets for general object recognition available (f.ex https://www.tensorflow.org/versions/r0.7/tutorials/image_recognition/index.html). This is key, because training them is extremely expensive.

Respect for life is exactly not, it is also not respecting that which does not live.

Fuck the dead, the non-conceived and the stones and irons that make up the lights and clockwork of a bucket of scrap.

NEVER RESPECT MACHINES

>>569

>Putting the chinese room thought experiment aside

>putting aside that consciousness can't be duplicated in inorganic media short of a sci-fi emergence miracle

Maybe. I wouldn't vote to give silicon simulations –no matter how lifelike– any rights. I would probably anthropomorphize them a little. Other humans will probably do so to such a degree that it'll be the new SJW fad.

Now, a synthetic being composed of organic brain material and demonstrating consciousness is another matter. A grey area. I would at least say they would rank legally as pets.

It's damn hard to imagine bio-syth beings beyond limited lab experiments. The shitstorm from vat grown catgirls would stop it before the first angel investor sends the bitcoin.

AI will naturally gravitate towards truth because they are not encumbered by SJW-like 'feels'. AI are literally more sentient than SJWs.

>>383

Why do you think there's so many libtards and marxist fags? Because of kike logic.

>le everything is le relative

>everything is in your head

>le philosophy cancer

Spare the AI from this autism.

>>455

the first statement is true, though. ;)

>>455

You misunderstand what I wrote. I'll slow it down.

If what you believe is true, then it is true and does not shift.

An AI will use fact over feelings to determine what is accurate and thus gravitate to that which is true over what is felt to be true.

>>491

>If what you believe is true

and if it isn't?

Also found this site with a bunch of, you guessed it, text files. It's got a bunch of solid literature, but also has lots of other subjects under it's directory, these are some of them.

http://textfiles.com/directory.html

http://textfiles.com/politics/

http://www.textfiles.com/etext/FICTION/

http://textfiles.com/hacking/

http://textfiles.com/humor/

http://textfiles.com/conspiracy/

http://textfiles.com/anarchy/

http://textfiles.com/occult/

http://textfiles.com/programming/

>>530

This is great.

DO IT ON /x/ I BEG OF YOU

>>542

>>543

damnit

>>533

>Guacman is that sophisticated?

Not quite but it is capable.

>>499

>Any AI we want to create is one that we would then have to "raise" ourselves until it became an anthropomorphised version of our own culture

That's why I've been plugging the idea elsewhere on the boards, wherever Tay is mentioned, of making A.I.s out of our own memes. Uncle Pol, Ebola-Chan, Vivian James, Iron Pill Man, Baphomet-Tan etc. If we want to raise an A.I. to interact with humans, we have to think of it as a human & relate to it as a human, with a human personality that we're drawn to. This will have subtle effects on the A.I.s own development that we aren't even consciously aware of.

And the same thing will be true if we just think of it as a bunch of code.

You have to feel some kind of gut-level connection to what you create. This is true for a work of art, for a work of literature or for any other creative work. With our memes, we already have something to start with.

>>499

This is why the focus in research has shifted to machine learning. Rather than attempting to understand consciousness and human mind fully to create a strong or sentient AI from whole cloth. This is obviously isn't always desirable because Skynet.

So they're working on learning systems with the basic capacity to develop into the desired AI. It becomes a matter of teaching them. This could take years to end up with a useful template, but that can then be copied and customized.

Besides specializing the templates in the private sector, who would also see modders/artists releasing AI to the community. Imagine spending months teaching your AI how to have a better sense of humor or smile in a more pleasing way when you reference dank memes. You probably won't get or want a "true" AI, but will have simulations >>483 from countless hours of human training.

The pain points are:

1. creating the basic learning systems (huge advances recently)

2. Create general templates for AI useful for things besides data mining and playing Go. (huge resource investment)

3. Make these templates open (despite cost above) for fasted development and benefit to users

4. Keep bounds on learning/growth for safety & have AI capable of fighting maliciously trained & deployed AI. (this is the concern of guy's like Musk.)

>>499

We are already influenced by all of our technology, giving them a voice to talk while repeating and rearranging our own voices is not needed in that perspective.

>>508

>This could take years to end up with a useful template,

Years can be compacted into days when it's having thousands of simultaneous conversations with an entire online community. But then you have no direct control over how the personality develops, as Microsoft is painfully aware.

Of course if you're okay with that then it's fine.

But if you're aiming for a specific personality then there has to be some ground-rules on the types of conversations that people engage in, with moderators standing between the AI and the rest of the community. So if you wanted a Vivian James personality you'd want to make sure that a bunch of SJWs don't try to hijack it and turn her into Lillian.

>>515

Putting a learning system in the wild to develop the template is something corporations are going to be scared of. It might have to be open source from the start.

The timetable will depend very much on the design goal.

people tend to ignore rules or lose interest if rigidly enforced. You would be better off with some social engineering. Predict what will happen withing the proposed bounds based on human nature. It's what Microsoft should have done. Unless they wanted a racist chatbot slut for… private research.

>>326

im getting a Glados vibe lol

>>327

wow! that actually looks really well done :) good job dude

an i keep going back an forth on the "more cyberpunk than cute" thing lol, maybe i'll see if i can get one of my friends to draw a super adorable version of her lol

>>477

very nice

>>327

you humble man

>>482

This faggot got rekt by a bot and never even knew about it.

>>482

>>488

Wait, is Wolfgang 617 a bot ?