has lee sedol killed himself yet?
Why would he? He's dreaming about spending more time with AlphaGo-tan
>>53545232
No, he is still playing go and fucking bitches.
I want to enter her database
>>53545232
Who?
>>53545299
she doesn't have a database tho
>>53545232
Koreans don't commit sudoku, they only know taekwondo
>>53545232
>>53545299
Lewd.
bump for alpha-go chan, anyone have more pics?
>>53545314
and how to win on StarCraft :D
>>53545232
>known for only one thing
>only for it to be stripped away by a machine
>>53545699
Yeah everyone will refer to the fuckhuge cluster that is alphago as the best go player when asked. You're dumb.
>>53545537
>people post dumb fanart instead of reading & discussing the alphago paper that can be found via google for free
I like anime a lot, but you guys are seriously missing the point here.
I'm killing myself because I can't do long division faster than a calculator
>>53545299
Hey, wait a minute, what happened? The data is all gone!
>>53545232
Why compete with machines when you can use them to build an Utopia for yourself & friends?
Plebes cannot into computation graphs
>>53545798
because utopia is a flawed concept.
>>53545743
nobody cares you fucking nerd
>>53545851
Nah, its only in your head. Solving most of the human problems is quite possible given energy and material resources we have. What we really lack is some entity that will use its intelligence and attention to tirelessly solve our problems for us, one after another. Humans don't do it because they are small, weak and uncoordinated. We can't coordinate on the global scale and that has prevented us from building a real utopia, even though we have enough science & technology.
Lee Sedol was killed by Alpha Go before this series.
The man we saw play was actually just a body double. Alpha Go plays the long game.
Please stop talking about convnets and post more moe pictures of alphago chan. Im a stanford ml grad student and even I dont give a fuck about your papers. stop
>>53545914
No, population will increase when we have more resources so resources will always be scarce. And trying to maximize any set of rules to this inevitably leads to a dystopia instead.
>>53545928
No, his body is controlled by the google botnet
alpha go only lost once to make it more plausible
>>53545903
Not caring is for plebes.
>>53545314
Actually South Korea is #2 in "The Highest Suicide Rates Worldwide"
>>53545947
Of course (mostly) consensual population control methods will have to be devised. Still, even with all these small restrictions the quality of life in the machine-governed utopia can be much higher (and perfectly uniform) than even in the developed world. Imagine not worrying about food, housing, health, crimes etc and have the freedom to live your life as you wish. That's what Demi's Hadsabis probably has mind when he says that Deepmind's mission is to solve intelligence and use it to solve everything else.
>>53545943
Afraid of anons teaching themselves some ML and outcompeting you, Mr Ivy? These ML frameworks are quite easy to use.
>>53545914
A boy is playing a ball. Dumb ass robot
Utopia literally translates to no place btw
Why? He got $120K or something like that.
>>53550460
It's not about what he got, but what he lost.
>>53545232
What changed from
>go is the hardest game, a computer will never be able to play it effectively
To this?
I hope it wasn't pure computational power but actually smart design
LSD is my hero.
He caused a supercomputer to shit itself.
>>53550577
It was basically brute forcing.
>>53550577
Google was able to simulate human intuition.
>>53550577
Original Go programs just brute force all moves.
Google's Go program brute forces only moves human players are likely to make. The AI provides information, given a Go board state, which moves on that board a human player is likely to make. This reduces the pool of moves to be brute forced tremendously - but when an actual human player makes a move that AI predicted as very unlikely, he might be able to catch the machine off-guard.
>>53550577
Advances in computing theory -- Neural networks. The closest thing we have to behaving like a brain. Eg. A blank slate brain gets trained into something -- This is the output file..the weighted brain. The brain with a bias. ie. a brain trained to say jump when it see's red, because thats the bias of the weights to jump instead of xyz.
This combined with genetic algorithms.
Computers allow 1000's if not millions or more of generations to be 'born' a second.
Sample genetic pseduo code:
Breed(1000);
TestAllOnTaskX();
EvaluateFitness(); // ie highest game score, longest alive etc
GetBest(50);
ShareGenticAttributes+5%mutation();
Repeat;
SO making neural networks with genetic algorithms is ..basically what biology has been doing forever and what we're doing now in computers way,way,way faster per second.
>>53550732
Funny how your post is paragraphs and paragraphs in length and does not even answer the question properly.
>>53550732
Your pseudo-intellectualism is showing, and it's showing too much you dumb faggot.
>>53550750
Yes it does... are you illiterate?
Question was:
>What changed from
>go is the hardest game, a computer will never be able to play it effectively
To this?
I hope it wasn't pure computational power but actually smart design
>>53550732
Answer explains directly 'what it was' which allowed it to go from something a computer can't play effectively to something a computer can be the best at. Even with pseudo code examples of how it's achieved.
>>53550845
1. Neural networks have existed for more than 50 years.
2. Google's Deep Mind uses such network in its basic form, which existed long long ago.
3. Your code has no relation to how alphago is implemented. The only thing AI does, as specified by paper, is it shows on a static board, which moves are more or less likely to be made by a human player.
There's no breed, no share attributes. It's a simple neural network with back propagation, as shown in Deep Mind promotional materials.
>>53550617
no you moron the time it would take to calculate every possible move doesn't get any less rediculous because we've had an advance of 20% in computational power
>>53550732
>faster per second
>>53551054
He could have made a simple direct sentence or two,
but no. He is better than that. He is an enlightened person, enlightened by his own intelligence.
He must strive to look the best, to look like he's at his very best at every single thing he does.
To look stylish, to look like his fedora.
So damned be those simple speech patterns of normal human beings.
We must adorn our replies with as many big words as possible, we must add code into our speech because code makes us look enlightened! We must add words that don't make sense in our sentences and statements, for they do in fact make sense. They make the sentences look more intelligent!
>>53550028
where did you find this
>>53551286
A fedora wouldn't have said "way, way, way"
You went way too far on a tangent
>>53551054
>implying that was his first crime against english
>>53545976
This is only because of the high rate of fan-deaths
>>53548004
>Utopia is always flawed
Its a stupid meme.
>>53550732
Deepmind does not use genetic programming though. They use convents in alphago, perhaps with some random search for hyperparameter tuning.
You are right in something though: in theory given enough time, the GA will find the best solution because it is a global optimizer.
>>53550894
Convents are from 80-90s, so more recent. Also many tricks that make then work well were found only recently.
>>53552885
>>Convents are from 80-90s, so more recent. Also many tricks that make then work well were found only recently.
But the core model is still the network of neurons, that could have been implemented long ago, albeit less efficiently.
>>53550577
>speculating instead of just reading the paper
Don't be a plebe https://gogameguru.com/i/2016/03/deepmind-mastering-go.pdf
>>53552898
Well you are right in a broad senseabout the core model, but there were so much architectural innovations, new optimization algorithms, that it leads to qualitative ly new performance levels. Superhuman visual pattern recognition has become reality in some domains: http://people.idsia.ch/~juergen/superhumanpatternrecognition.html
>>53553008
But as I see it, what's more important in case of Alphago is that google's guys applied neural networks to Go in a way that noone else thought of before.
