AlphaGo vs Ke Jie, Game Two

No one seems to realize that the moment Ke Jie loses he will be branded as a disgrace and traitor for China, effectively destroying its life

>>60571379
What the fuck are you talking about, shut up.

Stream is live now.

>>60571379
He's already lost to it 3 times now, retard

>>60572340
Those were not serious games.

>>60571247
>inb4 china gets their hands on goolge IP because some googler fucked it up, somehow allowing the chinks to get close to their hardware/software

File: 000174.jpg (9KB, 200x240px) Image search: [Google]

9KB, 200x240px

based michael

File: kidding_me.jpg (57KB, 565x541px) Image search: [Google]

57KB, 565x541px

>>60571247
>No eyecandy
mfw

>>60571247

I'm sad they didn't live stream the actual conference yesterday that had all the juicy technical details presented on the new AlphaGo.

Guess I'll have to wait for the paper to appear in Nature or something.

>>60572515
My understanding was that they didn't give too much new info.

It's about to begin shortly. We must endure the horrible piano tune they thought was a good idea.

#autism

GET IN HERE. starting in 30 secs

File: 1184833258.jpg (212KB, 1920x1080px) Image search: [Google]

212KB, 1920x1080px

WE ARE A GO

>>60572617
they had better music when alphago played lee sedol

File: 1354900275.jpg (218KB, 1920x1080px) Image search: [Google]

218KB, 1920x1080px

WHAT THE FUCK

ALPHAGO HAS NO FRUITS

THIS IS RIGGED

>>60572642
>the machine operator has shittier food

computer nerds cuked again

>>60572642
>the dude who just places alphago's piece also gets cheese and grapes

wew

JP stream for anyone who's interested.

https://www.youtube.com/watch?v=6vtuoMfdmjE

File: 1346200394.jpg (312KB, 1920x1080px) Image search: [Google]

312KB, 1920x1080px

NO SLUTS ALLOWED

File: alphago-3.jpg (104KB, 650x446px) Image search: [Google]

104KB, 650x446px

post alphaqt fan art

File: 1484347341500.jpg (84KB, 900x600px) Image search: [Google]

84KB, 900x600px

>"Yes AlphaGo only won by 0.5, but it was not at all a close game. It's an artifact of its training objective."

>it prefers to win by 0.5 with 99.9999999% chance instead of 10.0 with 99.99% chance.

This is why humans are so fucked. It just grinds you down and you have no fucking clue how you're doing. It just plays with you like a cat & mouse.

BTW, AlphaGo team runs on MacBook Pros. Pajeets with their ChinkPads BTFO!

>alphago going balls deep

IT'S ALL OVER

HOLY niggeR

WOOOOOWEEEEWOO WOO WOO WOO WOO

HI GANG

>>60572703
remind me never to put AI in charge of warfare

>well boys, we won the war!
>he says to the three 12 year old child soldiers with eight bullets left, in the middle of a scorched city

>"well it was a higher win probability bruh :^)

is this best of 3 or best of 5?

>>60572757
3

>>60572757
Neither. It's a three games series.

>>60572703
High-impact KEK @ yo post, nibba.

File: 1363188216.jpg (151KB, 1280x720px) Image search: [Google]

151KB, 1280x720px

White is already looking bad.

File: 1487116786649.jpg (139KB, 930x760px) Image search: [Google]

139KB, 930x760px

>>60572837
>White is already looking bad.
No shit. Lee SeDol will be the last human to win against AlphaGo.

File: graphic.png (11KB, 104x106px) Image search: [Google]

11KB, 104x106px

some random underpaid graphic designer is super proud about this minimalist reflection on the black piece

I guarantee it

Fucking Nigger dropped a stone and Michael bitched slapped Nigger's half bent body with a sharp order. Nigger is goofy.

>>60571247
who the fuck won game 1?

>white is building a wall to keep out black

PRAISE KEK

>>60572860
where is that stoned old guy from the time when alphago played lee sedol?

>>60572837
white fucked up.

hes letting black build up his BBC defence.

whitey dont stand a chance now.

>>60571247
Ke jie would win though if they doubled the board size.

>>60572868
alphago by .5 point

they are discussing why right now

>>60572922
>they are discussing why right now
this is why go is such a nigger-tier game.

leaving it open to chink interpretation is just asking to be cheated.

Male in black suit attempts to remain interested for hours with a peripheral of dread in the vicinity of the clock slowly ticking. Only so many "uhm-uh, right, right, sure, sure, okay" before the slow gnawing drowsy irritation peaks.

>>60572859
>underpaid

Minimalists don't deserve money, they're overpaid if anything. Literally MS Paint tier """art"""

File: chrome_2017-05-25_11-59-44.png (416KB, 730x389px) Image search: [Google]

416KB, 730x389px

NotLikeThis

File: 1461262274864.jpg (41KB, 688x456px) Image search: [Google]

41KB, 688x456px

>>60572971
>designers should be underpaid
good try shekelstein

File: BLACK SUIT NIGGER.png (75KB, 194x379px) Image search: [Google]

75KB, 194x379px

>>60572968
"LEAVE THOSE STONE THERE, NIGGER!"

-Michael 2017

I feel a sneaky move by ag coming soon

>>60573004
yup, it's just when the chink gets super into a local fight that master starts playing elsewhere on the board, then the chink is just its bitch to lead around

File: Screenshot from 2017-05-24 23-07-50.png (18KB, 143x109px) Image search: [Google]

18KB, 143x109px

"M-MICHAEL, I THINK niBBA MOVE STONE".

"Black not civil"
-Michael 2017

Kek Jie has lost his mind.

>>60572703
seems counter intuitive that by trying to win by 10 with a 99.99% chance, you make it so that you are less likely to win by 0.5 with a 99.99999999% chance

File: Screenshot from 2017-05-24 23-14-22.png (85KB, 216x375px) Image search: [Google]

85KB, 216x375px

"IM GET STONE so NOt to YELLING AT ME, michael"

>>60572997
>>60573049
>>60573111
Are you autistic?

File: JUST.jpg (103KB, 964x964px) Image search: [Google]

103KB, 964x964px

>>60573010
>>60573004
Looks like Ke Jie did the sneak move!

File: Screenshot from 2017-05-24 23-14-09.png (75KB, 181x347px) Image search: [Google]

75KB, 181x347px

"Michael, I thought he moved. s-sorry"

File: kejie.png (515KB, 650x588px) Image search: [Google]

515KB, 650x588px

HIS SECRET WEAPON!

>>60573108
perhaps it's more of letting a couple points go to "secure" a more secure victory? Like it beefs up the late-late game by giving away some easy points for the opponent

>>60573118
Kill yourself underage

Guy in black suit is fucking annoying with his empty comments.

File: eatingkejie.png (588KB, 666x720px) Image search: [Google]

588KB, 666x720px

>"yeah try doin this ya bitch ass robot"

>>60572997
>>60573049
>>60573111
>>60573140
>>60573154
>>60573210
Is your brain okay?

File: mpv-shot2146.jpg (197KB, 1920x1080px) Image search: [Google]

197KB, 1920x1080px

I wish I looked cool like Ke Jie

File: Screenshot from 2017-05-24 23-27-26.png (6KB, 163x116px) Image search: [Google]

6KB, 163x116px

>>60573222
>one brain
2

>>60573210
I didn't realize that was a kiwi he was eating.

Contributing to our alphago overlord with this captcha

how big is the computer that alphago is running on?

>>60573263
For you.

>>60573263
almost as big as my 14 inch BBC

Aww yes, cutie steph is comin in.

SLUTS ARE BACK

Nooooo where is based Michael going???

File: file.png (1MB, 982x768px) Image search: [Google]

1MB, 982x768px

>>60573263

Nice l-ladies... WEEEEEEWOOOOOOOo WOO WOO WOOO WOO

HI GANG

>>60573295
That's pretty big 4 me

>>60573299
No, /ck/ nooo....

>>60573263

It's on a single machine in the Google Cloud that includes the 2nd generation of TPUs, or the Google Tensor Processor units.

It's not distributed anymore like how AlphaGo was when it played Lee Sedol, which you can see at >>60573295. They mentioned this during Match 1's post press conference.

File: 1493237648442.jpg (36KB, 680x471px) Image search: [Google]

36KB, 680x471px

>>60573288
4chan, classy as always.

File: TPU5_2.jpg (786KB, 4964x3913px) Image search: [Google]

786KB, 4964x3913px

This is what AlphaGo runs on. Just one of these.

These smart ladies could keep me as a sex slave and I would obey to this best of my abilities. I'm a rag. Nice game so far, boys.

I'd rather watch alphago vs alphago

File: tpu-V2-hero.width-1000.png (680KB, 1000x692px) Image search: [Google]

680KB, 1000x692px

>>60573381
That is the old one. This is the new one that alphago is using.

File: tpu_cityscape_forwebonly_final.jpg (539KB, 2000x868px) Image search: [Google]

539KB, 2000x868px

>>60573412
Side view.

>>60573450
those are some thin-ass capacitors

like wafers

>>60573450
imagine these are alphagos boobies

>>60573450
I'm assuming you cannot put a system spec on that?

>>60573477
a few of them have more petaflops than the worlds strongest super computer for its specialized task.

>>60573312
>>60573295
wow...

>>60573412
is google using these for things besides alphago?

>>60573474
please don't discriminate against the dimensionally challenged

>>60573486
Fuck.

>>60573381
>wood screws
Holy fucking shit. 3.5 tflops confirmed.

>>60573490
making truck-pig drawings

....no I'm not kidding. Well, not only truck-pigs, but that was part of it https://research.googleblog.com/2017/04/teaching-machines-to-draw.html

>>60573501
They're also 8-bit operations IIRC. They're better at doing lots of parallel computations at lower accuracy, but for AI 8 bits is plenty.

File: machine_learning.png (28KB, 371x439px) Image search: [Google]

28KB, 371x439px

>>60573515
...are we just glorified piles of linear algebra? those drawings look like the drawings i have on my fridge that were made by my niece.

>>60573545
>...are we just glorified piles of linear algebra?
we dont know what the brain uses.

nobody knows how the neuron works or what it does just yet.

if someone finds out you can bet they wont be saying anything.

File: Screenshot_2017-05-25-15-55-37-397_com.google.android.youtube.png (1MB, 1920x1080px) Image search: [Google]

1MB, 1920x1080px

The hosts are just playing their own game on the other board

>>60573536
I think the new ones are 16-bit floating point and the old ones used 8-bit integers.

>>60573552
>>60573545
we know how the brain "works" as in the chemistry and which part does what. We also know which part *does* "what", but we don't how HOW it does it all.


by the way, ever wonder why foot fetishes are so common? it's a combination of brain plasticity and the fact that the section responsible for the feet is awfully close to the pleasure section....

>>60573552
>if someone finds out you can bet they wont be saying anything

Because you can signal the end of the world if you can control a go playing computer.

>>60573593
>we know how the brain "works" as in the chemistry and which part does what. We also know which part *does* "what", but we don't how HOW it does it all.
we dont actually.

neurons internal structure is unknown as its too small and complicated to see into the nucleus of it.

>>60573159

Alpha Go uses a type of search tree neural network, so it's very possible that winning with 99.9999% is on a separate branch.

>>60573628
If they're using a search tree, the standard algorithm is "minimax" where it doesn't matter if there's the possibility of a 10-point win, if the opponent has a way to block that chance.

I'm curious if the algorithm doesn't actually use a pure minimax algorithm and is actually doing some kind of probability thing, in which case it might have a reason to go after the larger lead

But all this is irrelevant late-game where you don't really have a reason to accumulate more points than it takes to win.

>>60573578
this.

wtf are these dumb bitches doing? do they even care what moves are getting played?

>>60573610
what? yeah we do. Motor neurons are well understood. The axon/terminal system is understood. The myelin whatever is understood. We know about the refractory period and the action potential and reuptake with all those SSRI's and stuff... we know the individual stuff pretty well

File: 5007db01d195b38c62892ac17ff8b54a.jpg (32KB, 512x443px) Image search: [Google]

32KB, 512x443px

>>60573610

Second >>60573703
We know neurons very well, and we have methods to visualize the signals propagating even up to a few neurons at a time.

The problem is the fuckton of connections between neurons, which are way harder to map and make sense of

File: peyote.jpg (123KB, 800x600px) Image search: [Google]

123KB, 800x600px

Ke Jie needs to drop his ego and take more time with his moves

>>60573578
This is why the beginning of the game is always the most interesting. After the start they just seem to do whatever the fuck they want on the board.

>>60573745
plus, the weirdness of markers and how the create memories somehow, and how everything is based only on quantity of signals rather than the "quality' of individual signals

brain is weird, yo

>>60573764
been a long time since i danced with mescalito

>>60573703
>>60573745
I definitely disagree with both of you.

theres a reason you will never see a picture of its internal structure and the exact working of the nucleus.

>>60573450
Google's opening up a few TPUs for researchers on their cloud.

On one had, I'm getting more and more scared of google every day, but on the other, I really really want to play with one.

>>60573578
they're showing you implied sequences. this is very standard if you watch go regularly

File: 18216599_10155182371865011_2621452465764143299_o.jpg (164KB, 1536x2048px) Image search: [Google]

164KB, 1536x2048px

>>60573798
You're always welcome back, friend

>>60573784
>how the create memories somehow
this scares me. i'm pretty sure something like 1 neuron dies in your brain per second, but we don't lose memories at a rate of 1 per second, which means memories are distributed. but damaging the brain doesn't destroy logical "groups" of memories, it just destroys random memories.

you would think that if you destroyed a few hundred thousand neurons responsible for holding memories, you would start to lose specific parts of your memory. for example damaging one part might destroy some of your memories of your mother, destroying another might damage some of your memories of your 10th birthday, and so on.

but no, if you destroy parts of the brain... random memories get destroyed. there's no logic to it. the brain just does what it wants.

i'm not even going to bring up those weird people who learned fucking Russian after being in a car accident.

sticky when

>>60573545
>...are we just glorified piles of linear algebra?
you'd be surprised how many things are linear algebra if you allow abstract non-numeric types and operators

graphs (complex networks, i.e. social networks, and neural networks, and the brain) are in most cases reducible to abstract algebras. graphs model an enormous number of physical and conceptual cases, so by extension...

File: 13064_2007_Article_24_Fig1_HTML.jpg (527KB, 1200x1436px) Image search: [Google]

527KB, 1200x1436px

>>60573802
here's the structure of a contralaterally projecting serotonin-immunoreactive deutocerebral neuron

>>60573881
thats not inside the neuron you retard.

>>60573848
well, what I've read is that for some memories they are competently decentralized. Like people loose half their brain and they don't forget much at all.

>>60573848
it makes sense if you understand hash functions and their relationship with information theory

the short version is that in a lot of cases, if you need to encode something, it makes a lot more sense to map it to some random value and bucket it, than it does to organize it in a principled hierarchy. hierarchies are expensive, "random" mappings are fast

>>60573892
um, yes? it's taken with a confocal microscope and they labeled everything with antibodies. hardly any computer manipulation. This shit is just a handful of μm across; it's hard to snap photos of it.

>difficurt

>>60573967
>This shit is just a handful of μm across; it's hard to snap photos of it.
isn't that literally what he was arguing?

>>60573967
>it's taken with a confocal microscope and they labeled everything with antibodies.
so you have no idea what you're looking at in reality.

its an incredibly small part of it that you dont even know the details of.

THE BOYS ARE BACK IN TOWN

https://www.youtube.com/watch?v=hQo1HIcSVtg

THE BOYS ARE BACK IN TOWN

https://www.youtube.com/watch?v=hQo1HIcSVtg

THE BOYS ARE BACK IN TOWN

https://www.youtube.com/watch?v=hQo1HIcSVtg

>>60573988
it is what I was arguing, this guys a complete retard.

Based Michael is back

>>60573988
I don't know much about cell microscopy, but modern microscopes can go to nanometer resolution super easy, and you can get sum-micrometer resolution with light alone. If there's a problem, it's not smallness

>>60573108
>seems counter intuitive that by trying to win by 10 with a 99.99% chance, you make it so that you are less likely to win by 0.5 with a 99.99999999% chance
i'll try to explain it as a go player

local and global states on a go board have a "temperature". high temperature states are very uncertain, low temperature states are very certain. but high temperature states are usually associated with greater potential value, because they are more variable. one of the most important and difficult skills in Go is choosing between potential value and certain value

once AlphaGo sees how it is going to win, it naturally tries to avoid hot board states. it actually wants the board to cool as fast as possible so that there's less potential variance, and more certainty that it has won

top-tier players do this, too. pro games usually end with a tight margin of victory

Finally Michael. I enjoy hearing him speak while watching Stephanie. Too bad they switched the pairings.

Godammit I want to see Michael's scenario but they aren't showing it

File: 1426858342702.jpg (31KB, 456x320px) Image search: [Google]

31KB, 456x320px

>Ke Jie keeps messing up his own hair

JUST

>>60574097
I miss steph

Does alphago keep a running prediction of how likely it is that it will win? Because that would be an interesting number to see

>>60574139
Yes, as it always picks the move that gives it the highest chance of winning

Will Lee Sedol or Ke Jie be the John Henry Irons of Go?

>>60574139
Yes. There's a control room where the DeepMind monitor the algo

>>60573848
How would you know if you'd lost a memory?

> Godammit these bitches didn't say anything useful and now I've gotta catch you all up and this retard keeps interrupting me

File: sadfrog.jpg (100KB, 1000x746px) Image search: [Google]

100KB, 1000x746px

>>60574172
when you're looking over old tests you took in your college days and have no idea what your answers mean

>>60574172
The same way you realize you've forgotten something.

>>60574172
>sit around thinking of the time you lost your virginity or something
>recall various things
>realize you don't remember what color your gf's hair was
>ohshit.mysql

>>60574172
write all of them down.

if you dont remember some then you lost it.

>>60574172
i believe that even amnesia patients that have literally no long term memory can still realize they've lost their memory, they just have trouble remembering that for more than 2-3 minutes

Why is this so interesting to watch? I know literally nothing about Go

>>60574288
I'm just watching the reactions of people who actually understands these shit.

>>60574288
desu one reason is that go is actually incredible interesting. it's a very different kind of board game than e.g. chess

>>60574316
It's also the last game that humans were better than computers at

>>60574325
besides starcraft and many other video games

I found an old facebook post from 2015

"Deciding the best computer go program should be made with playing also with the top proffesionals.. they play 4 handicap with old proffesionals.. let them play with iyama yuta or lee sedol.. ke jie.. shi yue.. ichiriky ryo... :)"

>>60574325
Quake, CS, etc.

>>60574351
I'm pretty sure an aimbot would have no problem with those...

File: Screenshot_20170525_080418.png (353KB, 730x709px) Image search: [Google]

353KB, 730x709px

Ke is having a heartattack.

>>60574362
he's in love... WITH A MACHINE!

>>60574325
games are almost invariably things that humans are bad at. they're "intellectual" because they require slow, inefficient, conscious reasoning. from a computational perspective, walking is much more impressive than playing Go. i don't mean to diminish the achievement but i want to put into perspective; soccer is computationally harder assuming the rules require 2 legs

File: ch.jpg (99KB, 1920x1080px) Image search: [Google]

99KB, 1920x1080px

>>60574362
He knows he's fucked.

I like that they're constantly building up this big ko battle that's eventually going to collapse into one glorious cascade that wins everything

File: Screenshot_20170525_080558.png (90KB, 302x324px) Image search: [Google]

90KB, 302x324px

It's like I'm looking into a mirror.

Doesn't this faggot know that you're supposed to undo your suit buttons when sitting?

>>60574361
By that train of thought then an AI could also win at Starcraft because having APM that fast you could make yourself invulnerable.

>>60574383
give the kid a break, he's only 19

>>60574362
>>60574376
>AlphaGo is sitting around twiddling it's digital thumbs and stroking it's digital dick while Ke has an existential crisis
how can humans even compete?

are we sure that AlphaGo isn't doing this on purpose? has it adjusted to scare the human opponent to achieve victory?

File: RWDS.gif (297KB, 500x281px) Image search: [Google]

297KB, 500x281px

>>60574387
NO FUCK YOU FAGGOT, COPY PASTING FROM THE LAST THREAD

It seems like most people in this thread are claiming that high APM and the ability to have perfect control over everything happening on the map will give the AI an advantage that will force a win.
Here is a video of one of the best current StarCraft bots losing to an D-rank (low skill) human player. The bot's APM is ~5500 while the human's is ~200. https://www.youtube.com/watch?v=ztNYOnx_YQo [Embed]
The fact is, no AI has ever beaten even an amateur player in a tournament. Even with great micro, if your play is too predictable then the human will learn it and exploit it.
I, for one, am very excited to see the development of new StarCraft AIs. And especially SC2 AIs so that it can challenge the current world champions.

>>60574332
I don't know the state of modern starcraft AI, but if you haven't seen the Berkeley Overmind, it's pretty interesting.

>>60574387
It did end up with something like 10k APM, doing some in-and-out attacks with mutalisks. But the devs balanced mutalisks around human APM, so of course a computer will be able to pull of some amount of bullshit.

>>60574375
which is crazy because human neocortex can readjust their roles if say vision is gone from birth.

if you could control what a human brain learns, which can learn complex motor control like playing soccer, imagine how good a brain training solely on playing a game like go would be

>>60574397
>And especially SC2 AIs so that it can challenge the current world champions.
sc2 is dogshit and my iphone could beat the world champs.

>>60574397
That was his point

File: kek.jpg (25KB, 326x326px) Image search: [Google]

25KB, 326x326px

>>60574403
>I don't know the state of modern starcraft AI, but if you haven't seen the Berkeley Overmind, it's pretty interesting

Don't you remember the Human VS AI matches that took place with Overmind right after the tournament that year?
Watch an amateur human tear through its mutalisks. https://youtu.be/hmXY_pkU2Wc?list=PL9F69781768E85502&t=591

WE SHALL HAVE KO

>>60574409
no his point was wrong because

1. aimbots CAN beat pros TODAY
2. AIs with high apm CANNOT beat amateurs TODAY

>>60572746
ai does whatever it is programmed to do, it is not a characteristic of all ai

>>60574397
I'm just saying that if you allow the bots to have unlimited APM then they do bullshit like this
https://www.youtube.com/watch?v=IKVFZ28ybQs

>>60574397
Maybe once google is done with Go they'll move on to some esports games. It would be fun to see a SCII / League AI, and I think that if they've beaten go, they certainly can beat those.

Certainly at some point we'll get a general game-playing AI, which would be cool and might unlock a whole new class of games

>>60574439
i would love a war thunder AI. you could replace the shitty bots in that with proper bots.

could we get EU4 AI?

>>60574433
>1. aimbots CAN beat pros TODAY
eh, not entirely. I remember watching a match of CS pros in MM vs bots and I think they won using mollies and nades from areas where they couldn't get shot easily.

>>60574435
I'm fairly certain that video is a gimmick. How do the zerglings know which one the tanks will target? Anyway, a human could just use something else to fight zerglings if it could do this.

Also, DeepMind is focused on AIs that LEARN to play. Not ones that are pre-programmed to do cheeses like this. So an AI learning how to do strategies like this would be extremely impressive by itself since SC2 has such a large search space.

>>60574457
>Anyway, a human could just use something else to fight zerglings if it could do this.
There's many different types of micro bots though. If google just made an AI that strung them together it'd probably be pretty good.

File: 1441864806963.png (72KB, 500x482px) Image search: [Google]

72KB, 500x482px

>>60574435
That's pretty neat

It would be very cool to see Google tackle some video games, but unfortunately the only game I really care about (Melee) is all about exploiting human deficiencies and there already exist bots that are effectively invincible

DESIGNATED
waiting period.

>>60574490
i wanna see a Magic:tG bot that includes the deckbuilding meta

>>60574457
You generally train these bots against themselves at least a little, and as long as there's some marginal benefit to some strategy it will likely find it. The thing that kills AI is a false local minimum.

So if an AI messed with the position of its zerglings a little and found that it could save a few by moving them away from the tank splash, you can bet it would try moving more of them further, and other such micro techniques.

>>60574506
pls no, it would just reinforce the meta even harder as it learns how to beat anything non-meta efficiently.

All of these games that people are talking about that they'd like to see, have a much more complex ruleset than go.

Usually I'd say that AI has an advantage when there are more rules and less nuance, but we haven't seen that case yet -- go is maybe the simplest, most nuanced game we have.

This guy plays with his hair almost as much as I do. I can't tell if he thinks he's fucked or if he's getting excited

>>60574545
>This guy plays with his hair almost as much as I do.
JUST

>>60574568
He's been in JUST mode for quite a while now.

0 - 3

rip ke jie

I wanna fug steph

>>60574641
i would ko my white stones all over her

So what's happening?
the bot already win?

>>60574641
I wanna watch Steph suck AlphaGo's superior BBC after it 3-0s Ke Jie

I kinda want to see Ke Jie when he's not getting fucked by an AI. It seems like here he's (justifiably) killing himself over every move.

>>60572515
(((They))) are hiding obviously.

is it over?

>>60574680
The chigga resigned

Ke Jie just resigned to give AlphaGo a 1.5 - 0 lead

>>60574680
yes, black won, white resigned

Poor chink

>>60574692
do they keep playing?

humans are obsolete.

So can AlphaGo form a proper conversation unlike the current meme bot that just spam overused line

i gotta tell you that was honestly a "man who knows his limitations refuses to challenge christopher robin" kind of resignation

>>60574710
no, but there will be another match in a few days

>>60574698
poor guy just keep looking at the table
it's like my anime

>>60574717
will literally never happen. go is a deterministic game which can just be brute forced. you can't codify language like that

Wtf is google doing, why they continue to show this beaten chink???
REEEEEEEE fucking white piggu and their AI

>>60574733
go provably can't be brute forced, it's in EXP-C. alphago is heuristic, it doesn't "solve" go

i'm certain any language decision problem you would ever propose is in a formally difficult class, and yet we successfully reason about language heuristically both as humans and in modern ML

File: 1453610433531.gif (582KB, 512x288px) Image search: [Google]

582KB, 512x288px

This is making me sad. Ke Jie looks like he's fucking dying inside.

>>60574752
>go provably can't be brute forced, it's in EXP-C
thats not a proof dumbass

>>60574733
The whole point about go is that it can't be brute forced. Each move has 200 posibilities, try to search that tree too deep and you very quickly exhaust the computational ability of the universe

>>60574718
Did anyone ever clear that game? I doubt it. Never seen any screenshots of the ending.

>>60574762
>the computational ability of the universe
imagine being this brainlet to believe this has been found in any rigorous way.

so who is better: Ke Jie or Lee Se Dol?

>>60574755
it's proven, AI is godlike in whatever it do, we have created god, unlike what the bible said

>>60574782
Ke Jie

>>60574762
There are infinite number of universes, anon. The only thing limited is your imagination.

>>60574757
EXP-C != P by the time hierarchy theorem, that's as strong a proof as you're ever going to get out of complexity theory

>>60574792
cool but you didnt prove that it cant be brute forced by all methods.

>>60574718
>>60574771
What game?

>>60574778
there's actually some work on theoretical maximum computational density. it is finite because you can't have infinite information in a finite volume, and cross-volume communication is bounded by the speed of light

it's obviously not a useful bound, though

>>60574813
>basing anything on the standard model
into the trash it goes

>>60574811
Winnie the Pooh Home Run Derby

>>60574806
it can be brute forced, but if you try to store all of the information you need to brute force it, you will quickly run out of universe to store the information in

this isn't even a "well someone who doesn't really like all of this newfangled stuff is deliberately misconstruing information to dissuade us from doing something new" thing, this is a "if we can do this all of our knowledge of the way the universe works is false" thing.

>>60574838
>but if you try to store all of the information you need to brute force it, you will quickly run out of universe to store the information in
proof pls?

>>60574787
we created the bible and the god defined in it as well

>>60574806
>>60574813
>>60574838
if you really want to butt heads against this shit here's a good entry point:
https://en.wikipedia.org/wiki/Limits_of_computation

GG humans are obsolete after WW3 : the thread

Just waiting on them NK NUKES

File: Ke Jie.png (742KB, 1312x738px) Image search: [Google]

742KB, 1312x738px

>>60574755

>>60574787
>first thing we teach it to do is dispirit the smartest among us

>>60574850
>using the standard model
into the trash it goes

>>60574838
>but if you try to store all of the information you need to brute force it, you will quickly run out of universe to store the information in
You limiting factor then is time, not space.

>>60574841
>Tromp and Farnebäck showed that on a 19×19 board, about 1.2% of board positions are legal (no stones without liberties exist on the board), which makes for 3361×0.01196... = 2.08168199382... ×10^170 legal positions "of which we can expect all digits to be correct" (i.e. because the convergence is so fast).[9] It has been estimated that the observable universe contains around 10^80 atoms, far fewer than the number of possible legal positions. As the board gets larger, the percentage of the positions that are legal decreases.

>>60574858
>hold information to figure out what parts of the game you have already brute forced
>whoops you just filled up the entire universe AGAIN

>>60574755
>>60574787
>>60574852
Before AlphaGo, Ke Jie always appeared cocky as shit in his videos and interviews.

>>60574841
Not the guy you're discussing with but 200 possibilities per move and you compound that 200 times and make pathways thats a hugefuck shitload tree for a program to traverse if you ask me. Stop being a pain in the ass the dude is not trying to be a smartass nobody's trying to hurt your ego.

>>60574866
>It has been estimated that the observable universe contains around 10^80 atoms, far fewer than the number of possible legal position
imagine being so brainlet you use individual atoms to store information.

fuck off brainlet.

>>60574876
>retard is wrong
>gets called out on it
>LOL STOP BEING BUTTHURT
good try retard

>>60574880
>using 10^80 atoms to store 10^170 legal positions
we can't even store a bit without using less than like 4 atoms right now

>>60574891
>because im too stupid to store more information, that makes it impossible
good try brainlet

>>60574876
when you're standard for proof involves solving all of physics then you're quite fucked since it's well known there are logically true statements that cannot be proven

honest to god, for computer scientists saying something is provably in P is equivalent to saying it has no formal optimal solution, except for a few exotic examples which are definitely not in EXP-C

File: Screenshot_20170525-135929.png (334KB, 1440x2560px) Image search: [Google]

334KB, 1440x2560px

>>60574888
Sorry not samefagging. Im not an expert in these things but i get the feeling that this game is not bruteforcable. Heck i never even looked up the rules before 2 days ago when the first game started.

>>60574913
>but i get the feeling that this game is not bruteforcable.
>feeling
I cant even begin to explain how retarded you are

>>60574912
*provably not in P, obviously

almost all of CS is impossibility proofs, it's a defining aspect of the field. AlphaGo is very much based on those assumptions, you can't say something dumbass like "AlphaGo uses brute force" and then reject all the models on which AlphaGo is actually based. you're rejecting the science, by that logic AlphaGo is a mystery that shouldn't work

>>60574919
Sorry if I'm calling out the retard who keeps asking for proofs like an autist when the other guy is just trying to have a discussion. Try social life. You might get better at holding up conversations.
>>60574850
Dont bother with the guy seriously

>((4/3 * pi * (radius of universe ^3)) / (planck length ^3) * (time until heat death / planck time)
10 ^ 338

>number of go games with 400 or fewer moves
10 ^ 800

>>60574944
yeah i know he's a dumbass who is engaging in the scientific equivalent of "are we there yet," you can always demand more proof. Godel's thereom shows this is fruitless. i'm sure he'll call me a dumbass which is fucking hysterical since they're handing me my PhD tomorrow

Anyone who knows about the TPU, what do you think of the direction AI accelerator chips are going?

I've been getting interested in neuromorphic chips that have architectures inspired by biology and the brain, but it seems the direction Google is going is just parallel processing on traditional architectures.

Is this just because the research isn't there yet, or do we trust that Google knows best and accept the status quo?

Where the fuck is the press conference?

File: Screenshot_20170525_093000.png (91KB, 640x446px) Image search: [Google]

91KB, 640x446px

PRESS CONFERENCE HYPE

>>60574782
Lee actually won a match.

>>60575151
>against AlphaGo which would lose even with 3 handicap stones to the current one

This is the worst translator I've ever heard

>>60575162
Not AlphaGo's fault that it took Ke Jie so long to fight him.

>>60574435
Honestly, since the zerg is supposed to be a hivemind and shit shouldn't that be their actual behaviour?

>>60575148
Wait, are they just talking about the game, or will they talk about AlphaGo as well?

>>60575163
Good god this woman is so useless I'd rather just hear the Chinese

>>60574782
>>60575151
>>60575162
Even though Lee Sedol would get swept by AlphaGo's current incarnation, he'll always be remembered as the more based player for making it bleed prior to its ascension to an unstoppable, invincible god.

https://www.youtube.com/watch?v=kJy37GtkcF4

Alphago should give Ke handicap for the last game.

File: 1367980161870.gif (1MB, 250x250px) Image search: [Google]

1MB, 250x250px

TFW that at this very moment Google Deepmind is creating a Starcraft II AI with its APM intentionally limited to human levels so that it can crush humans with strategy and planning alone, no cheesy APM bs

https://deepmind.com/blog/deepmind-and-blizzard-release-starcraft-ii-ai-research-environment/

"Computers are capable of extremely fast control, but that doesn’t necessarily demonstrate intelligence, so agents must interact with the game within limits of human dexterity in terms of “Actions Per Minute”"

>>60574490
How does it feel to care deeply about a game so brainless that even dumb bots can be better than you at?

File: 1image.jpg (14KB, 365x247px) Image search: [Google]

14KB, 365x247px

>>60574490
>>60575241

I spoke with the guy who wrote this paper:https://arxiv.org/abs/1702.06230

Video:https://www.youtube.com/watch?v=dXJUlqBsZtE

They are now creating a melee AI which is going to be limited to human levels

>>60575195
This is actually pretty cool. From what I've seen, at high level there's a lot of Go where a single move decides the match, and a lot of very famous single moves.

>>60575265
Meant to add that its APM and reactions time will be limited, so that the AI must outplay its opponents instead of cheesing them with ridiculous APM

>>60574490
This would be true for all fighting games, to be honest. In real time, any match might look spontaneous and unpredictable but at any given frame there's an optimal input for that situation.

If you let a player slow down a match such that he can experience it frame by frame, he could study every available option's hitbox and frame speed data. So given as much time as he needed, he could pick the best input for every frame.

Something like AlphaGo could trivially do that same thing but in real time.

>>60575222
>tell AI she can only so 300 APM
>AI strategizes on where to place those 300 actions whithin that minute
>suddenly three outbursts of 100 APms in a minute

anyway, am I the only one who wants to see AI trying Portal?

>>60575288
They're going to limit the reaction time & APM of the AI in this paper... see here:

>>60575283
>>60575265

>>60575293
>AI
>In 3 dimensions
I want to see one of this AIs in any 3d game.

Ooh, they're taking questions now.

>>60575311
>I want to see one of this AIs in any 3d game.
They have neural nets playing quake

Oh, interesting. Apparently they think AG is worse at complicated situations, which is counter to what I'd expect. Announcers were saying that it was a bad move for Ke Jie to add complexity, but I guess he knows better than they did.

File: thinkingTime.png (51KB, 853x457px) Image search: [Google]

51KB, 853x457px

heh nothing personnel kid

What the fuck was that?

>>60575370
>a computer
>thinking
More like computation time.

raise question in real CHINESE WAY ONREEEEEEEEEEEEEE

Programmers are saying it was even for the first 100 moves. Maybe Ke Jie would do better if they gave him time to stop and think for a bit?

>>60575394
>pedantic cock muncher has to reply to a humorous post

Do you think that Google forced the alphago to play first 100 moves aiming to reach 50% instead or 100% to make the game appear close to flatter chinks?

>>60575394
It's up to you to prove that thinking is something more than computation.

nice microphone rape by this "translator"

>>60575411
It's patronizing to compare a computer to the human brain. They're non-comparable.
>>60575428
>what is conscious thought
AlphaGo AI is trained, there is no consciousness to speak of.

>>60575496
>They're non-comparable.
Why? Are you a special quantum mechanical snowflake?

>Ask question only in THE REAL CHINESE WAY REEEEEEEEEEE

lmao this was great

>>60573848
It's a subtle thing, and that's why after a time parts of a memory change, it becomes more polished as each time you remember it it's like a little story/picture you recreate for yourself and your reinforce the parts that gets a stong response for you.
In those inumerable neurons dying you probably do lose parts of memory but you probably don't even realize it because they're not the important part of the memory.
It's also why people's testimonies about details aren't worth much.

>>60573241
Terry, is that you?

>>60575496
Again, you act like you know what thinking actually is. You don't. You are the one patronizing here. It is incorrect to claim that thinking is more than just computation until you can present considerable evidence of that, which one one has yet.

>>60575577
Cognitive thought and consciousness cannot be quantified by science nor computers, which is evidence enough. Anything that operates on instruction sets does not think, and the brain does not function so. Neural networks are just computational models, and don't come close to quantifying cognitive thought.

I hope you all are remembering the basilisk now

>>60575613
The fact that we can't is not a proof nor evidence.

>>60575613
>I can't define consciousness, so tough luck science can never quantify it
ok.jpeg

File: 1479446048078.jpg (145KB, 633x758px) Image search: [Google]

145KB, 633x758px

>tfw chinese diaspora and seeing this chinese autism
idk why they're sucking china's dick so much here. Are they being paid?

>>60575677
>>60575712
We don't even know if human brains get close to being conscious and cognitive.

>>60575293
>am I the only one who wants to see AI trying Portal?
http://www.youtube.com/watch?v=W9gxFkOz2_4

>>60575421
>Do you think that Google forced the alphago to play first 100 moves aiming to reach 50% instead or 100% to make the game appear close to flatter chinks?
No it's reaching that score due to its algorithm which maximises the chance of winning rather than the score.

>>60575744
No idea, they have absolutely no reason to do so as Google is treated like shit in China.

>>60573412
>>60573381
AlphaGo runs on a fucking cluster of these, not one.

>>60575962

It's probably something like 20, assuming the biggest server motherboard and the most IO available to Xeon where it runs 20 x4 PCIE lanes which would make sense where the TPUs are possibly running in a 4x interface.

They improved AlphaGo to the point where they don't need a distributed computing cluster to make it work anymore though as DeepMind has mentioned.

>>60575915
Yes, but as google's engineer commented, and unlike all other games alphago played previously, internally it evaluated the game to even for the first 100 turns. This can be explained by the human playing extremely well, or by machine, instead of trying to win, trying to make the game appear even for a number of turns, and afterwards switching to its usual win condition.

File: 198384.jpg (11KB, 171x266px) Image search: [Google]

11KB, 171x266px

Accelerationist future when?

File: Screenshot_20170525_113138.png (39KB, 589x186px) Image search: [Google]

39KB, 589x186px

>>60575962
No, Google has specifically said that this AlphaGo can run on a single chip.

>>60576150
If that's the case then AI is going to be absolutely everywhere.

>>60576158
Using a trained neural network doesn't take much processing power. The training itself is the hard part and requires weeks or months of processing even with racks and racks of ASICs like these.

>>60576201
How much would one of these cards cost? Wouldn't these cards be competing with Nvidia for the autonomous car market?

>>60572365

The goal post. It moves thusly.

Is this the same bullshit like Deep Blue where Kasparov was vs. the machine AND people? He did not truly complete against the AI because people behind the scenes were helping the AI make more human-like choices.

>>60573412
I can't wait until Google announces that the TPU version 4, or 5 or 6 ... was not engineered/designer d by humans anymore but an AI, at least partially.

>>60576252
Self-driving cars rely pretty much completely on heuristic methods, not neural networks. Google's chips are built specifically for computing neural network stuff.

File: kasparov.jpg (44KB, 638x412px) Image search: [Google]

44KB, 638x412px

>>60576280
I remember that event, Kasparov literally could not comprehend that a computer could out "think" him.

Also, no I don't think they are feeding this it any moves.

>>60576252
Google doesn't sell them.
You cannot get it.
You need to lease it from their TPU (((Cloud)))
https://google.ai/

>>60576309
>not neural networks
not even from the computer vision? Thanks for answering my questions. Are neural-nets worse than heuristic methods in this area?

File: kasparov0.jpg (520KB, 1920x1249px) Image search: [Google]

520KB, 1920x1249px

>>60576318
another picture of his emotional state

>>60576327
well that's a bummer

>>60576361
Wait couple months and buy a cheap chinese knock-off.

File: 220px-St_5-xband-antenna.jpg (21KB, 220x282px) Image search: [Google]

21KB, 220x282px

>>60576305
like this antenna that was made by an AI doing space searching?

>>60576318
That's the thing though. I don't believe Deep Blue won fairly. There have been plenty of discussions about IBM doing some shady things behind the scenes guiding the AI to victory. I mean, Kasparov lost, but the idea that he lost purely to a machine is most probably not true. Also as far as I know IBM either dismantled Deep Blue or locked it away somewhere shortly after this event which is very peculiar and makes you wonder why they aren't interested in a rematch or playing other chess grandmasters.

>>60575329
What?

>>60573864
>if you allow abstract non-numeric types and operators

What the fuck, that's not linear algebra then. Linear algebra is about vector spaces of number fields you shit. Graphs are topological objects (1-CW-complexes)

>>60576442
I agree that the circumstances surrounding it were suspect but only a few years later there was no man who could beat a computer at chess, and many of these programs were not running on a cluster but on pcs.

https://en.wikipedia.org/wiki/Computer_chess

>>60576456
https://gym.openai.com/envs

>>60576280
There is no human strong enough to help AlphaGo

>>60576569
that's the thing, isn't it? when putting it up against the best, what good is human intervention? they're only going to make things worse

File: 1485101722084.jpg (182KB, 730x864px) Image search: [Google]

182KB, 730x864px

Just like poetry.

>>60575311
There is one learning how to drive a car in GTA 5 right now.

>>60576150
Look at Deep Blue.
It used to take a supercomputer to run it..
Now you can have a better version of it on your freaking $200 phone.

File: AlphaGo-Lee-Sedol-game-3-Lee-Sedol-enters-venue.jpg (56KB, 744x608px) Image search: [Google]

56KB, 744x608px

>>60576615
kek

>>60571379
China already censored these matches from any live stream/TV in PRC.

>>60576701
You gotta be kidding.

>>60576478
I'm not that anon and I don't know what he was talking about, but
>Linear algebra is about vector spaces of number fields
They can be any fields, not only number fields.