Life In 19x19

The moment when Cho Chikun approved of Lee Sedol (Legend Match 2)
https://youtu.be/CMICkMHbJ1M

[Baduk Lecture 7] Must Know Variation of Invasion 3 (Very Large Knight Move Attachment)
https://youtu.be/JYIu3gsjExQ

[Tygem 9 Dan Baduk Live Review 2] The way to win a losing game
https://youtu.be/98JM5Ghi4kw

BadukDoctor wrote:The moment when Cho Chikun approved of Lee Sedol (Legend Match 2)
https://youtu.be/CMICkMHbJ1M

Great tesuji! I did a quick check to see if my bots would find it: LeelaZero 254 didn't within a minute, but the old KataGo that came with Lizzie 0.7 did in just a few seconds! Interestingly that's the other way round to another (more standard) 1st line tesuji in one of my games (84 in viewtopic.php?p=250468#p250468) where LZ found the tesuji but KataGo didn't.

Uberdude wrote: I did a quick check to see if my bots would find it: LeelaZero 254 didn't within a minute, but the old KataGo that came with Lizzie 0.7 did in just a few seconds!

I tried twice with 15-block LeelaZero that came with Lizzie 0.7. After 50k playouts it didn't find the tesuji the first time, but LZ found it the second time after a few thousand playouts. So maybe you were unlucky with LZ254 and lucky with KataGo (and maybe not, but in any case trying just once doesn't prove that KG is better at finding the tesuji than LZ).

I attach the file, in case someone would like to check (or to follow the lecture while replaying variations).

I came here to post about A7, but I see it's already been discussed. That move is incredible. I paused the video when he said it was coming up, and I couldn't imagine anything like it. It's not the type of move that's in my search space...

I'm still in shock.

The God of Endgame - Lee Chang-ho [Legend Match 3]
https://youtu.be/dUK6BWxQN8Q

jlt wrote:

Uberdude wrote: I did a quick check to see if my bots would find it: LeelaZero 254 didn't within a minute, but the old KataGo that came with Lizzie 0.7 did in just a few seconds!

I tried twice with 15-block LeelaZero that came with Lizzie 0.7. After 50k playouts it didn't find the tesuji the first time, but LZ found it the second time after a few thousand playouts. So maybe you were unlucky with LZ254 and lucky with KataGo (and maybe not, but in any case trying just once doesn't prove that KG is better at finding the tesuji than LZ).

I attach the file, in case someone would like to check (or to follow the lecture while replaying variations).

Mine is Lizzie as well. It found out that move in less than one second. Is it related to the display card? I run this in my another computer with cheap display card and it took much longer for Lizzie to found out.

a7 was in my search space, mostly because I paused the video for a minute specifically searching for something interesting to play. Actually, the move feels relatively normal to me, as far as tesuji go. I think if I still had comfortable enough main time in a game, I’d probably find it, just like I find a million other crazy attacks that don’t work. Then I’d be somewhat surprised when it doesn’t immediately end the game in one direction or another. What’s the saying, that an idiot’s mistakes and a genius’ masterpieces look identical?

But really, finding the candidate move just noticing that b6 and b8 are desirable places to cut, and making effective use of all the aji instead of just half of it. It’s a very western-chess style move. And once you find it, it’s not that hard to read out.

Sometimes I wonder whether putting effort into learning go as a study is really just putting effort into demystifying the magic. It’s fun to be able to find moves like this, but it was even more fun when I couldn’t find them and just had to let my jaw drop and marvel. Well, I guess it’s been the most fun to play tesuji in my own games.

Off to the next video!

BadukDoctor wrote: Mine is Lizzie as well. It found out that move in less than one second. Is it related to the display card? I run this in my another computer with cheap display card and it took much longer for Lizzie to found out.

I think that once the computer decides to look at the move, it realizes immediately that it's a good move, however this happens after a random number of playouts (somewhere between 2000 and 200000?). Of course if your computer has a good graphics card, it reaches 2000 -- 200000 playouts in fewer seconds than the computer with a bad one.

On the other hand, unlike Lee Sedol, LZ cannot find the tesuji 24 moves in advance

jlt wrote: I think that once the computer decides to look at the move, it realizes immediately that it's a good move, however this happens after a random number of playouts (somewhere between 2000 and 200000?). Of course if your computer has a good graphics card, it reaches 2000 -- 200000 playouts in fewer seconds than the computer with a bad one.

Weird, I've just tried a few times (restarting to make sure I'm getting different random number seeds), and LZ consistently identifies A7 at the best move within 5,000 playouts or less. I've tested with networks 157, 242, 258 and "bubbles-15" (the default with Lizzie installation).

Note that the policy network value for A7 is only 0.1%, so it should be a "hard to find" move. But the high policy moves all favour black by a large margin, meaning that LZ will search more broadly. So A7 gets a single playout quite early in the search. (Tip: in your LIzzie settings, change "min playout ratio for stats" to 0.01 so that you can see the moves that it glances at and rejects quickly.) However, the initial evaluation of that early playout could be low; it might take a while before LZ gets around to searching that move more deeply. And this can be affected by "blind spots" in some of the variations. Maybe it's similar to this example.

[Baduk Lecture 8] The difference between 3rd line and 4th line
https://youtu.be/afAlusHUWqw

[Pro Match Review] Oh Yujin's Aggressive Play (Cho Chikun - Oh Yujin)
https://youtu.be/Uyi1k8gJyRs

The reason why strong players don't play knight move (All About Aji 2)
https://youtu.be/4ySexrPZBt4

Getting into 3.3 after Opponent Defends the Corner with Knight Move (All about corner 4)
https://youtu.be/X343XBVaD9k