Yeah just to clarify, no captures. The highlighted black move is move 35.
Fun fact: AlphaGo Teach positions run AlphaGo only on the last 20 moves of each variation. Since this variation is 35 moves long, the first 15 moves should be pre-AlphaGo human moves.
1 Like
Only 10 white stones instead of 17 on purpose?
Yes, it feels completely unnatural to me for everything to be all the way to the right, that’s the best I can do without just clicking wherever 
1 Like
Wait, then how did they get the winning percentages for the moves they didnt run AlphaGo on…? I realize only the last 20 moves of each variation are “original”, but I assumed the previous moves would still have receiced 10 million playouts each.
@mark5000 just meant that AlphaGo didn’t play the first 15 moves (it did analyze them, but that doesn’t matter for us here trying to guess patterns)
It’s an important observation, since it means we might see some old human joseki 
2 Likes
2 Likes
Gonna just wait for yebellz
My sequence for round 7? this should be round 7 instead of 6 right?, Gia’s was #6
Summary
P.S. notice some numbering missing, and added comments about my thought process.
I like low Chinese opening, this one sided black group looks just like some kind of Chinese opening concentrated so much on the right. And white must have a pretty good left side and influence to balance this thick black territory placement.
The split the board in diagonal half I’ve seen and played some myself, no problem with those starts. The trouble is why there is a thick connected black shape around P7 P9, it implies white put some pressure to them from the outside. And that leave the strange ponnuki shape on the upper right. White must have sacrifice some stones to exchange for the influence. At first I thought some capture must have happened, or sabaki kind of play to prompt black makes the ponnnuki and double tigher’s mouth. Then I noticed that given white still has such high winning percentage, must means there are some moves AI likes so much (besides 3-3), so I feel a shoulder hit like K16 is pretty typical and famous from AlphaGo, it must love to play it and putting pressure to the upper left black group.
2 Likes
I think I found the position in the AlphaGo Teach tool
1 Like
Well, I was still waiting for yebellz to post a guess, but congrats, you found it…
2 Likes
Thanks. I have to say that I’m impressed by how close @claire_yang and @mark5000 came without searching the Teaching tool.
So we have…
Gia:+4 -8 = -4 points
Groin: +14 -11 = 3 points
le_4TC: +20 -7 = 13 points
mark5000: +30 -2 = 28 points
claire_yang: +32 -1 = 31 points
(gennan: cheated ;P)
claire_yang only got a single stone misplaced by a single line! Pretty impressive.
6 Likes
I assumed that the hint you gave about the teaching tool meant that we were allowed to use that! 
Honestly, I still think it’s pretty impressive that you found the position, even on its head
It’s one of 47 positions in AGT’s database after 10. L3: Play Go at online-go.com! | OGS
I am curious, how do you search through all the variations? looking for them one-by-one? Or do you have a rough idea like I used has an assumption that it’s a type of Chinese opening, along with common corner joseki then only looking through those? If that’s the case, still have a good sense of the direction in order to reduce that.
I had a hunch that it evolved from a low Chinese opening and after that, it was almost a one way street in the Teaching tool.
1 Like
I am surprise the stone I am fairly certain it’s the AlphaGo AI move turned up to be the wrong one. Those outside influence to pushdown P7 P9 group, I am actually not very sure, I went with them simply due to existing black stones have to be placed there. I personally won’t attach and push all the way from the outside at P10, it’s very risky black has a chance to make it into a running fight toward the center and keep getting bigger upper side territory.
I’m also not sure the lower left corner white enclosure. Just happened to pick the simplest one, due to previous guessing games I tried to be fancy but their answers ended up with just the simple ones. I am actually quite surprise AI doesn’t use different enclosures.