This article is about endgame technique which Viktor Lin 6 dan EGF played against Viacheslav Kaymin 5 dan EGF in Starrise Grand Prix Stage 1.
There is not much to talk about, I hope the reader can find the endgame technique before scrolling down to answers and analysis. I did deep analysis on 3 AI’s as well. Leela Zero 40 blocks 195 – 8b28fe3c and A1/ 9A of Pangafu Master series.
Can you find black’s next move? #1
Black played the marked move (which is wrong), how can white punish? #2
After black captured, for white to throw in is the correct endgame technique, this gains white several points. How?
After black captures, white can play the second beautiful sacrifice tesuji.
After black takes with 2, white throws in once again at 1 and black takes. We can see that white managed to reduce black’s eye space.
White plays atari to force black to connect. If black connect, white will win, so black is forced to play 2 and to fight for this ko. But then white takes first and this is gain for white. The game went a bit different and white couldn’t manage to continue well and he lost by 1,5. Still, he found this endgame technique, now lets count and then take a look on how AI manages to find it.
Normally black has 6 points if black connects which is the correct move to protect against white throw-ins. However, if white throws in and then throws in once again, and throws in for last time, black will have only 4 points and white will be able to possibly gain 1 point. Although my counting is not quite correct (never) I guess it is about 3,5 points comparing to 6 points. Can we say that the throw in is worth 2,5 points? ( If I am wrong, please correct me, learning is endless process).
With close to 600 000 visits, Leela 40 blocks is convinced that black has to defend, although it found this move a lot before, it is convinced that it is working even close to 600 000 visits. So definitely Leela 40 blocks manages this problem.
If you capture as black did in the game, LZ will not play the throw in on lower side but instead it wants to throw in on left. They are pretty much same, it is anyway forced both ways. It does not matter much.
The same test on 9A of Master series of Pangafu . We can see that it failed and it couldn’t see the defending move. Sometimes AI’s like to play some other moves and then return to position, but in this case it is not this way. It even after exchanging moves on top, it does not want to defend.
Once you play some of the moves it wants, it finally wants to play this descend on right side, and then after you wait a bit, it figures out that white can throw-in. The percentage is only going up.
A1 also does not manage to find the move. I let it think until 800 000 visits.
After black takes, it doesn’t want to throw-in, even after you play some of the blue moves in top right, it still doesn’t want. My expectation was that AI trained with some human games would have easier time finding the endgame technique, but it didn’t hmm. In any case this yose technique is very intriguing and I will try to make more articles about yose, it is so deep.
Below you can find the entire game.Follow me in social media: