|
Yooo getting back into the game. Can someone toss me an invite to the channel? I'm 'ceezy' = on KGS
|
#
?
Mar 13, 2016 09:09
|
|
|
|
| # ? May 20, 2024 19:38 |
|
|
Added you
|
|
#
?
Mar 13, 2016 09:18
|
|
|
The resign
|
|
#
?
Mar 13, 2016 09:44
|
|
|
Humanity starts the fight back!
|
|
#
?
Mar 13, 2016 09:45
|
|
|
Well that was unexpected! I really want to see some analysis so we can see where it all went wrong. Or right if you're Lee Sedol. What would the sentiment on team DeepMind be? Personally I think (or hope) that'd I'd be excited. This sort of thing could be exactly what's needed to put an upper bound on its strength and maybe it shows where there's still margin for improvement. But I dunno, maybe having it play more games against itself would be even more valuable. It can play a lot more games that way at least.
|
|
#
?
Mar 13, 2016 10:03
|
|
|
The beginning of the comeback was when Alphago responded wrongly to the L11 wedge that Lee Sedol played in the top middle.Demis Hassabis on Twitter posted:Lee Sedol is playing brilliantly! #AlphaGo thought it was doing well, but got confused on move 87. We are in trouble now... Demis Hassabis on Twitter posted:Mistake was on move 79, but #AlphaGo only came to that realisation on around move 87 Demis Hassabis on Twitter posted:Lee Sedol wins game 4!!! Congratulations! He was too good for us today and pressured #AlphaGo into a mistake that it couldn’t recover from
|
|
#
?
Mar 13, 2016 10:55
|
|
|
for those with lots of games on ogs, this is kinda neat http://vatcss.info/games/gotstats/ i win way more often as white for whatever reason. maybe cause i get white when playing weaker opponents? i suppose that makes sense.
|
|
#
?
Mar 14, 2016 22:13
|
|
|
totino boy posted:for those with lots of games on ogs, this is kinda neat Ditto. I mean I'm a small sample size candidate, but going 10-2 as white and 14-22 as black is a pretty stark difference. It also said my most triumphant victory was against Shadonra (4d) in an intense game. Meanwhile in reality Shadonra was on pace to beat me by like 100+ until he lost due to time out (I think he timed out of all the 2015 goontourney games simultaneously).
|
|
#
?
Mar 14, 2016 22:21
|
|
|
Pander posted:Ditto. I mean I'm a small sample size candidate, but going 10-2 as white and 14-22 as black is a pretty stark difference. I think the silliest thing that I've happen is winning by a margin greater than 81 in a 9x9 game (Japanese rules of course).
|
|
#
?
Mar 15, 2016 04:14
|
|
|
Game 5: White has 1 group; black has 4. No wonder B is going to lose
|
|
#
?
Mar 15, 2016 09:52
|
|
|
Hi I'm Dancer I'm a total loser with no Go experience, and I made a KGS account with the same name. Could someone grant me access to ITGO?
|
|
#
?
Mar 15, 2016 16:18
|
|
|
Can I get an invite to ITGO? pmrowla on KGS
|
|
#
?
Mar 15, 2016 16:31
|
|
|
Well, while Lee Sedol lost last night it was by a very thin margin, capping off a really entertaining series. I have been following all the matches, even falling asleep watching them late into the night. I think Game 4 was won due to the Horizon Effect and that humans only barely have the ability to put a move outside of the purview of AI. Since AlphaGo's AI will be "unfrozen" and it will be allowed to learn from the 5 matches against Lee Sedol, I highly doubt any human will be able to take advantage of this effect ever again. I would still like to see a match with Ke Jie, if only to shut his arrogance down. Some people are obviously despondent about this historic loss for humans and while I'd rather avoid the sticky philosophical issues in this thread, I will say that this despondency is just silly. Pretty much every online Go server has seen a massive surge in registrations. News outlets like Time, ABC, the NY Times etc etc are writing at length about Go. When's the last time Time magazine has written about Go? We seem to project onto computers a rigidity of thought, but AlphaGo is teaching us to cast away this prejudice. In our losses we let that prejudice reflect back on us, but even in this early stages of computer dominance, we've already seen the humans adapt because of AlphaGo's dominating play. Fan Hui has said since he played AlphaGo that he is now a much better player. And indeed, due to tournament wins he's risen about 300 spots in the world rankings. And I think we've watched Lee Sedol change and adapt his tactics. Would we have ever seen a move like White 78 in Game 4 if Lee Sedol wasn't under so much pressure to find a way to beat AlphaGo?
|
|
#
?
Mar 15, 2016 16:50
|
|
|
Welp, just won a game against a sandbagger by 0.5 points because he forgot to take a dead group of mine off the board. I did as well, but he was the first to hit done and I didn't bother to look it over. Serves him right. There's really no way someone who has once attained the rank of 2k is suddenly now a 17k player.
|
|
#
?
Mar 15, 2016 18:36
|
|
|
Overwined posted:Welp, just won a game against a sandbagger by 0.5 points because he forgot to take a dead group of mine off the board. I did as well, but he was the first to hit done and I didn't bother to look it over. Serves him right. There's really no way someone who has once attained the rank of 2k is suddenly now a 17k player. There was this guy on KGS who reached around 2k iirc but then suffered an accident and became a 15k or something like that. He explained this on his KGS profile.
|
|
#
?
Mar 15, 2016 18:55
|
|
|
Overwined posted:Well, while Lee Sedol lost last night it was by a very thin margin, capping off a really entertaining series. I have been following all the matches, even falling asleep watching them late into the night. I think Game 4 was won due to the Horizon Effect and that humans only barely have the ability to put a move outside of the purview of AI. Since AlphaGo's AI will be "unfrozen" and it will be allowed to learn from the 5 matches against Lee Sedol, I highly doubt any human will be able to take advantage of this effect ever again. I would still like to see a match with Ke Jie, if only to shut his arrogance down. same. agreed.
|
|
#
?
Mar 15, 2016 19:07
|
|
|
im excited to get back to playing go after watching it for 4 hours a day over the past week
|
|
#
?
Mar 15, 2016 19:09
|
|
|
totino boy posted:im excited to get back to playing go after watching it for 4 hours a day over the past week I've played 45 games since last Thursday 
|
|
#
?
Mar 15, 2016 19:19
|
|
|
Overwined posted:I would still like to see a match with Ke Jie, if only to shut his arrogance down.
|
|
#
?
Mar 15, 2016 19:29
|
|
|
If someone could invite me to ITGO (huzenhagen) that'd be great. Can't wait to get brutally owned as i work out how to play this game
|
|
#
?
Mar 15, 2016 19:31
|
|
|
Some factoids:
Xom fucked around with this message at 21:48 on Mar 15, 2016 |
|
#
?
Mar 15, 2016 21:37
|
|
|
Is Lee Sedol the last person who will ever win a game against AlphaGo?
|
|
#
?
Mar 15, 2016 21:42
|
|
|
Michael Redmond quoted that 1 in 10000 stat during last night's commentary. One of the AlphaGo developers tweeted that by move 89 (or something in there) black had realized that its response to White 78 was a mistake. I was curious about how AlphaGo expressed this, but we'll probably never know.
|
|
#
?
Mar 15, 2016 21:44
|
|
|
Overwined posted:Michael Redmond quoted that 1 in 10000 stat during last night's commentary. One of the AlphaGo developers tweeted that by move 89 (or something in there) black had realized that its response to White 78 was a mistake. I was curious about how AlphaGo expressed this, but we'll probably never know. 
|
|
#
?
Mar 15, 2016 21:48
|
|
|
such a good movie
|
|
#
?
Mar 15, 2016 21:57
|
|
|
So why was AlphaGo "frozen" for the duration of the series? A human opponent could potentially learn and adapt to a particular opponent's strategy...why handicap AlphaGo from being able to do that?
|
|
#
?
Mar 15, 2016 22:03
|
|
|
Alphago requires thousands of games to learn. Any games played against Lee would barely change anything.
|
|
#
?
Mar 15, 2016 22:27
|
|
|
jivjov posted:So why was AlphaGo "frozen" for the duration of the series? A human opponent could potentially learn and adapt to a particular opponent's strategy...why handicap AlphaGo from being able to do that? The software wasn't changing is what I assume they meant--they weren't tweaking any code but I assume it was still collecting the data from its games however insignificant it would be in the grand scheme of things
|
|
#
?
Mar 15, 2016 22:30
|
|
|
So apparantly, Ke Jie has challenged AlphaGo: http://weibo.com/u/2865101843?refer_flag=1001030101_&is_all=1#_rnd1458063222095 I really hope the AlphaGo team will accept this, watching these matches has been very interesting.
|
|
#
?
Mar 15, 2016 22:52
|
|
|
Rated PG-34 posted:Alphago requires thousands of games to learn. Any games played against Lee would barely change anything. gwrtheyrn posted:The software wasn't changing is what I assume they meant--they weren't tweaking any code but I assume it was still collecting the data from its games however insignificant it would be in the grand scheme of things So even "assimilating" or adapting to Lee's particular playstyle wouldn't help much or change much due to the sheer number of other Go information AlphaGo was drawing on?
|
|
#
?
Mar 15, 2016 23:11
|
|
|
All this alphago jazz got me to log into KGS again after not being on in quite a while. I've lost a fair bit of stongness but I am still more than capable to teach some beginners here and there if anyone needs some help.
|
|
#
?
Mar 15, 2016 23:26
|
|
|
jivjov posted:So even "assimilating" or adapting to Lee's particular playstyle wouldn't help much or change much due to the sheer number of other Go information AlphaGo was drawing on? That's basically what they said at one of the press conferences. It needs upwards of tens of thousands to make significant change. They said even feeding it all of Lee Sedol's game records wouldn't do an appreciable amount.
|
|
#
?
Mar 15, 2016 23:32
|
|
|
elementc posted:All this alphago jazz got me to log into KGS again after not being on in quite a while. I've lost a fair bit of stongness but I am still more than capable to teach some beginners here and there if anyone needs some help. I was just about to ask if anyone would be up for some 9x9s just to help me keep grasping the fundamentals...you available now?
|
|
#
?
Mar 15, 2016 23:32
|
|
|
Overwined posted:Michael Redmond quoted that 1 in 10000 stat during last night's commentary. One of the AlphaGo developers tweeted that by move 89 (or something in there) black had realized that its response to White 78 was a mistake. I was curious about how AlphaGo expressed this, but we'll probably never know. They mentioned that Alphago keeps a running estimate of its chance of winning the game. I think the developers tweeted that Alphago's estimated odds of winning barely budged on moves 79-88, but Alphago's estimated odds of winning crashed on move 89. So Alphago might not have "realized" exactly where the mistake was, but it didn't think it was losing until move 89.
|
|
#
?
Mar 15, 2016 23:32
|
|
|
mikeycp posted:That's basically what they said at one of the press conferences. It needs upwards of tens of thousands to make significant change. They said even feeding it all of Lee Sedol's game records wouldn't do an appreciable amount. Moreover it already has all of Lee Sedol's games up to the point he played AlphaGo. The real "learning" it will do will be at the hands of the developers as they analyze mistakes and fix the algorithms.
|
|
#
?
Mar 15, 2016 23:34
|
|
|
Overwined posted:Moreover it already has all of Lee Sedol's games up to the point he played AlphaGo. The real "learning" it will do will be at the hands of the developers as they analyze mistakes and fix the algorithms. It also has all of your matches, Ke Jie. AlphaGo is coming for you!!  I wonder if AlphaGo is "told" who it is playing so that it can emphasize analyses made on games played by that player as a human might.
|
|
#
?
Mar 15, 2016 23:36
|
|
|
golden bubble posted:They mentioned that Alphago keeps a running estimate of its chance of winning the game. I think the developers tweeted that Alphago's estimated odds of winning barely budged on moves 79-88, but Alphago's estimated odds of winning crashed on move 89. So Alphago might not have "realized" exactly where the mistake was, but it didn't think it was losing until move 89. I think that's really cool.
|
|
#
?
Mar 15, 2016 23:43
|
|
|
I thought I had done this at some point...but I guess not; can I get added to the ITGO list? Username is jivjov, same as here.
|
|
#
?
Mar 15, 2016 23:43
|
|
|
Overwined posted:Moreover it already has all of Lee Sedol's games up to the point he played AlphaGo. The real "learning" it will do will be at the hands of the developers as they analyze mistakes and fix the algorithms. As far as I know, it has 2 neural nets and a MCTS thing. Maybe something else as well but I forgot. I suppose they might be able to optimize some poo poo in the connections between everything, or maybe improve the MCTS a bit, but I think I read that they primarily still have a ways to go in improving the neural nets. They haven't reached the limit there. And that will just be done by having it play itself. So it doesn't really have any algorithms to fix. Not in the classical way. Overwined posted:It also has all of your matches, Ke Jie. AlphaGo is coming for you!! Doubt it. What I'm wondering though, is this: AlphaGo doesn't give a poo poo about maximizing score, only about winning. That's why it plays slack but safe moves when it's sure it'll win, and why it goes crazy when it thinks it's behind. I wonder if they could change at least the last one: if it doesn't see a way out, it could go for keeping the difference as close as possible, so when the opponent makes a mistake it could as least profit from it. This wouldn't really work against an computer opponent but it might against a human. Forcing a human to not make a single mistake in the endgame would make it stronger, maybe.
|
|
#
?
Mar 16, 2016 00:07
|
|
|
|
| # ? May 20, 2024 19:38 |
|
|
I'm about to put in an order with GoGameGuru and wanted to run it by the thread first in case I'm missing a way to get a slightly better deal...I have the Mini Me Go Set and the Cheap Go Board in my cart right now...This looks like the cheapest way to get a decent set of stones and bowls, plus access to 9x9, 13x13, and 19x19 board space, without resorting to a flimsy paper/plastic board. I'm just tired of using my thrift store $2.99 set that has stones that are smaller than dimes.
|
|
#
?
Mar 16, 2016 00:13
|
|