I'm guessing as soon as AI massively take away simple jobs Capitalism will collapse.
Edit: it's the politcs and economy turn in this thread i guess
| Forum Index > Dota 2 General |
|
Pontual
Brazil3038 Posts
December 07 2017 11:15 GMT
#183581
I'm guessing as soon as AI massively take away simple jobs Capitalism will collapse. Edit: it's the politcs and economy turn in this thread i guess | ||
|
PoulsenB
Poland7771 Posts
December 07 2017 11:27 GMT
#183582
great job Donald! | ||
|
PassiveAce
United States18076 Posts
December 07 2017 11:33 GMT
#183583
Love how every american president thinks that theyr gonna be the one to solve a problem thats very obviously gonna take at least half a century Non-reform jews suck so hard | ||
|
PoulsenB
Poland7771 Posts
December 07 2017 11:46 GMT
#183584
| ||
|
Pontual
Brazil3038 Posts
December 07 2017 11:59 GMT
#183585
| ||
|
PassiveAce
United States18076 Posts
December 07 2017 12:01 GMT
#183586
There are a bunch of evangelical christians who pushed trump to recognize jerusalem as the capital of israel because it will somehow speed up the apocalypse | ||
|
Faruko
Chile34173 Posts
December 07 2017 12:02 GMT
#183587
| ||
|
PoulsenB
Poland7771 Posts
December 07 2017 12:02 GMT
#183588
As for the American Christian extremists: "The shortest verse in the Bible is 'Jesus wept'. The only thing wrong with it is the past tense." | ||
|
SpiritoftheTunA
United States20903 Posts
December 07 2017 12:38 GMT
#183589
On December 07 2017 08:10 Nevuk wrote: Show nested quote + On December 07 2017 07:38 ahswtini wrote: man, august ames is dead, suspected suicide after she tweeted she didnt want to shoot porn with gay actors (for health reasons) I just looked her up and I don't really find her face very attractive same | ||
|
PoulsenB
Poland7771 Posts
December 07 2017 13:05 GMT
#183590
| ||
|
Slomo
Germany7198 Posts
December 07 2017 13:44 GMT
#183591
| ||
|
intotheheart
Canada33091 Posts
December 07 2017 15:26 GMT
#183592
What interested me was how the games looked very different from games you'd see between top GMs today, since those players prepare using opening theory and some soft rules with heuristics, while Alpha Zero just kinda played on its own and figured out what to do. I believe Alpha GO was trained the same way: playing itself rather than using some guidelines on the game. Just loaded it with rules, and let it go. | ||
|
Slomo
Germany7198 Posts
December 07 2017 15:52 GMT
#183593
On December 08 2017 00:26 intotheheart wrote: I'm watching a few of the games played between Stockfish and Alpha Zero. They're chess engines. Stockfish was (until recently) the best engine, having a rating of about 3400 (compared with the #1 in the world's 2800). Alpha Zero spent 4 hours learning chess,and just crushed it. What interested me was how the games looked very different from games you'd see between top GMs today, since those players prepare using opening theory and some soft rules with heuristics, while Alpha Zero just kinda played on its own and figured out what to do. I believe Alpha GO was trained the same way: playing itself rather than using some guidelines on the game. Just loaded it with rules, and let it go. No idea about the other networks you mentioned. But AphaGO trained on data that was gathered from human players to 'get the gist' (aka get really good) and only then it played against itself. The nice thing about this is that you can run hundreds of games at the same time and learning from all of them. | ||
|
intotheheart
Canada33091 Posts
December 07 2017 15:56 GMT
#183594
As far as calculations go, Stockfish made ~70million/sec, while AlphaZero only did 80000/sec. I believe it's because instead of analyzing every square or possible pattern, it only looked at a specific subset of them. This is pretty significant since it's not like it was a brute-force victory. | ||
|
Pontual
Brazil3038 Posts
December 07 2017 16:21 GMT
#183595
this is actually so cool Full of stuff that i would pay to try once and then wait until it's cheap to have it on day to day | ||
|
Sn0_Man
Tebellong44238 Posts
December 07 2017 16:26 GMT
#183596
The calculations comparison isn't really right. The numbers cited are merely the number of positions each program saw. It's worth noting that the systems were running on very very different kinds of hardware (AlphaZero ran on dedicated Neural-Network processors designed from the ground up to facilitate it's way of thinking) and a lot of comparisons indicate that while stockfish was on great hardware, AlphaZero was on insane hardware (although much less impressive than it's training hardware). Going back to the raw numerical comparisons, the whole point of machine learning games like this is to change how games are seen. It's been proven that you can't just solve chess by playing each game. There are more possible games than there are atoms in the universe. The problem is that modern chess engines essentially justify their judgement of a position based on the next position. Which is, in turn, rated based on the next possible position etc, but there are too many possibilities to properly solve out. AlphaZero, while obviously still assessing things based on possible future positions, is built to associate everything that happens with a chance of victory. It's able to understand a position without having to calculate a few million possible moves. | ||
|
Plansix
United States60190 Posts
December 07 2017 17:20 GMT
#183597
On December 07 2017 22:05 PoulsenB wrote: honestly the stories about sovereign citizens going to court are the best thing ever lul They are both my favorite and least favorite people. Their pleadings are hysterical, but also fill you with the dread that they are so unhinged that they might show up on your house. Highlights: Asking for payment in troy ounces of gold, to be measured in my weight. It is unclear why he decided my weight was the appropriate amount for damages. Asking for my attorney to be banished from the US. Telling a judge that they didn’t have authority over him, only to back down when contempt of court came up. Threaten a judge with a lawsuit in open court if the judge didn’t rule in his favor. Saying that the laws of the sea were the only laws that governed man. Arguing that the Magna Carta showed that the court had no jurisdiction over him. They are super fun people that practice moon law. | ||
|
intotheheart
Canada33091 Posts
December 07 2017 17:49 GMT
#183598
On December 08 2017 01:26 Sn0_Man wrote: If there's a "Zero" in the AI name, it means google never fed it human data (AlphaGo Zero, AlphaZero chess, etc). If there isn't, then it trained on human data sets first. AlphaGo (the famous one) was human trained. AlphaGo Zero eventually got to the point where it annihilated human-trained AlphaGo. The calculations comparison isn't really right. The numbers cited are merely the number of positions each program saw. It's worth noting that the systems were running on very very different kinds of hardware (AlphaZero ran on dedicated Neural-Network processors designed from the ground up to facilitate it's way of thinking) and a lot of comparisons indicate that while stockfish was on great hardware, AlphaZero was on insane hardware (although much less impressive than it's training hardware). Going back to the raw numerical comparisons, the whole point of machine learning games like this is to change how games are seen. It's been proven that you can't just solve chess by playing each game. There are more possible games than there are atoms in the universe. The problem is that modern chess engines essentially justify their judgement of a position based on the next position. Which is, in turn, rated based on the next possible position etc, but there are too many possibilities to properly solve out. AlphaZero, while obviously still assessing things based on possible future positions, is built to associate everything that happens with a chance of victory. It's able to understand a position without having to calculate a few million possible moves. I felt like highlighting the calculations as it is indicative of the difference in approach between the two engines. I'm curious how this changes the way top players approach the game, although it'd be kinda tough to adjust a lifetime of study and preferred patterns. | ||
|
Slomo
Germany7198 Posts
December 07 2017 18:14 GMT
#183599
On December 08 2017 02:49 intotheheart wrote: Show nested quote + On December 08 2017 01:26 Sn0_Man wrote: If there's a "Zero" in the AI name, it means google never fed it human data (AlphaGo Zero, AlphaZero chess, etc). If there isn't, then it trained on human data sets first. AlphaGo (the famous one) was human trained. AlphaGo Zero eventually got to the point where it annihilated human-trained AlphaGo. The calculations comparison isn't really right. The numbers cited are merely the number of positions each program saw. It's worth noting that the systems were running on very very different kinds of hardware (AlphaZero ran on dedicated Neural-Network processors designed from the ground up to facilitate it's way of thinking) and a lot of comparisons indicate that while stockfish was on great hardware, AlphaZero was on insane hardware (although much less impressive than it's training hardware). Going back to the raw numerical comparisons, the whole point of machine learning games like this is to change how games are seen. It's been proven that you can't just solve chess by playing each game. There are more possible games than there are atoms in the universe. The problem is that modern chess engines essentially justify their judgement of a position based on the next position. Which is, in turn, rated based on the next possible position etc, but there are too many possibilities to properly solve out. AlphaZero, while obviously still assessing things based on possible future positions, is built to associate everything that happens with a chance of victory. It's able to understand a position without having to calculate a few million possible moves. I felt like highlighting the calculations as it is indicative of the difference in approach between the two engines. I'm curious how this changes the way top players approach the game, although it'd be kinda tough to adjust a lifetime of study and preferred patterns. Well, yea. Neural networks and machine learning algorithms evolve amazingly fast. Google might publish something about their newest game AI today and everybody knows its a milestone from 6 months ago. Their current models are either better or more efficient. In the end its loads and loads of maths on how to make fancy matrix manipulation as efficient as possible. If you cant do it on software develop new chips to get the efficiency from hardware. | ||
|
andyrau
13015 Posts
December 07 2017 20:32 GMT
#183600
On December 07 2017 20:15 Pontual wrote: Capitalism is in it's late stage and we are close to a transition. What do you guys think about this? Will it happen in our lifetime? I'm guessing as soon as AI massively take away simple jobs Capitalism will collapse. Edit: it's the politcs and economy turn in this thread i guess interesting but probably not a direct causal relationship, and probably not for another 200+ years it's like saying the rise of modern farming techniques led to the downfall of feudalism and monarchies On December 07 2017 21:02 Faruko wrote: If only american presidents would not feel the need for total domination or actually believe they own the world this is akin to a child that thinks he/she doesn't need parental guidance On December 07 2017 21:01 PassiveAce wrote: That sounds amazing tell her to add me on fb There are a bunch of evangelical christians who pushed trump to recognize jerusalem as the capital of israel because it will somehow speed up the apocalypse i believe it On December 08 2017 01:26 Sn0_Man wrote: If there's a "Zero" in the AI name, it means google never fed it human data (AlphaGo Zero, AlphaZero chess, etc). If there isn't, then it trained on human data sets first. AlphaGo (the famous one) was human trained. AlphaGo Zero eventually got to the point where it annihilated human-trained AlphaGo. The calculations comparison isn't really right. The numbers cited are merely the number of positions each program saw. It's worth noting that the systems were running on very very different kinds of hardware (AlphaZero ran on dedicated Neural-Network processors designed from the ground up to facilitate it's way of thinking) and a lot of comparisons indicate that while stockfish was on great hardware, AlphaZero was on insane hardware (although much less impressive than it's training hardware). Going back to the raw numerical comparisons, the whole point of machine learning games like this is to change how games are seen. It's been proven that you can't just solve chess by playing each game. There are more possible games than there are atoms in the universe. The problem is that modern chess engines essentially justify their judgement of a position based on the next position. Which is, in turn, rated based on the next possible position etc, but there are too many possibilities to properly solve out. AlphaZero, while obviously still assessing things based on possible future positions, is built to associate everything that happens with a chance of victory. It's able to understand a position without having to calculate a few million possible moves. wasn't the margin 60-40 for alphago-zero vs alphago | ||
| ||
StarCraft 2 StarCraft: Brood War Britney Dota 2Shuttle Mini EffOrt BeSt Stork Rush FanTaSy Snow Leta [ Show more ] League of Legends Counter-Strike Other Games singsing3447 FrodaN1000 hiko929 B2W.Neo417 Liquid`VortiX142 ArmadaUGS121 QueenE101 KnowMe64 DeMusliM56 UpATreeSC30 Railgan10 Organizations StarCraft 2
StarCraft 2 StarCraft: Brood War Dota 2 League of Legends Other Games |
|
Replay Cast
KCM Race Survival
Ultimate Battle
ZerO vs Bisu
RSL Revival
BSL
eOnzErG vs StRyKeR
Cross vs JDConan
RSL Revival
IPSL
Ret vs ggnore13
nOmaD vs Aether
BSL
Bonyth vs Doodle
Kyrie vs Jimin
Replay Cast
WardiTV Weekly
[ Show More ] Kung Fu Cup
Solar vs Classic
ByuN vs SHIN
PiGosaur Cup
The PondCast
|
|
|