Elf 1 missed the best play in this position. Can you find it?
$$Wc White to play
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . O . . X . . . . . . . . . |
$$ | . X O O . . . X X O X X X . X X X . . |
$$ | . . X , O . O X O O . O . . . O X O . |
$$ | . X . X X X O O X O . . . . . . . . . |
$$ | . . O . . . O . X O . . X . . . O . . |
$$ | . . . . . O . . X . O O X . . O . . . |
$$ | . . X . . . O X X X O X X . X O . . . |
$$ | . . . . . . O O X O O X . O . X O . . |
$$ | . . . , . . O O O X X O . . . X O . . |
$$ | . . . . . . X O O X . O . . . X . . . |
$$ | . . O . . X . X X X O . . . . . X . . |
$$ | . . . . . X O . . X . . . . . . . . . |
$$ | . . . X . X O O O X . . . . . . . . . |
$$ | . . X O X X X O X O . O . . . . O . . |
$$ | . . O O O . X O X O . . . O . O . O . |
$$ | . . . O . O O X X O . . X X X X O X . |
$$ | . . . . . . X . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------
[go]$$Wc White to play
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . O . . X . . . . . . . . . |
$$ | . X O O . . . X X O X X X . X X X . . |
$$ | . . X , O . O X O O . O . . . O X O . |
$$ | . X . X X X O O X O . . . . . . . . . |
$$ | . . O . . . O . X O . . X . . . O . . |
$$ | . . . . . O . . X . O O X . . O . . . |
$$ | . . X . . . O X X X O X X . X O . . . |
$$ | . . . . . . O O X O O X . O . X O . . |
$$ | . . . , . . O O O X X O . . . X O . . |
$$ | . . . . . . X O O X . O . . . X . . . |
$$ | . . O . . X . X X X O . . . . . X . . |
$$ | . . . . . X O . . X . . . . . . . . . |
$$ | . . . X . X O O O X . . . . . . . . . |
$$ | . . X O X X X O X O . O . . . . O . . |
$$ | . . O O O . X O X O . . . O . O . O . |
$$ | . . . O . O O X X O . . X X X X O X . |
$$ | . . . . . . X . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]
Enjoy!
Last edited by Uberdude on Sat May 18, 2019 1:15 am, edited 1 time in total.
Reason:Added coordinates
The Adkins Principle: At some point, doesn't thinking have to go on?
— Winona Adkins
LeelaZero (should be about latest network) did not find the correct answer in 1.5 million playouts either, though it spent 6k playouts on the correct move at some point. It also thought white is losing. When I showed the correct answer, it took 50k playouts or so to see that it works and white has then a good lead.
I spent about 30 seconds looking in vain mentally at liberty situations in the wrong part of the board, then sat back just saw it.
Once I was in a car with with some friends and I spotted someone far ahead and said she's attractive. After we past my friends were amazed. "How did you know? Your vision is not that good!". I said, "Maybe, but a couple million years of human evolution can optimize for the right things. If it were a road sign I'd have had no chance."
Leon wrote:Hmm, I am quite suprised about the reports that Elf and Leela both miss the tesuji, as I expect an average 1 dan to find it in a serious game.
Is that local position an obstacle for them and can we find this blindspot in similar positions? What about other AI?
I think that the problem for self-taught AI is the rarity of the position. Given enough playouts, they could make the right play, but even Elf 1, given many playouts as a commentator, did not give the position enough playouts to choose the right play. Once the player in the game, Honinbo Satsugen Meijin, made the play, Elf recognized its value. But if you miss a play and your opponent, who is a slight variation of yourself, also misses it, it has to occur in many games before one of you catches on.
The Adkins Principle: At some point, doesn't thinking have to go on?
— Winona Adkins
I tried this position out on a few bots with playouts in the tens to hundred thousands. None of them found it as the best move though a few did devote some playouts to it. Once I played it minigo 990 was the fastest (a few seconds) to realise it worked (noticing that against the other black hane you had to do it again as mainline variation in Lizzie on hover). I would be interested if any of the old Monte Carlo bots (either with or without a policy network) could find it.
I would be interested if any of the old Monte Carlo bots (either with or without a policy network) could find it.
No
At least not Zen 7
thanks for the hint on sgf downloading of diagrams, I always forget and only click here to show code ...
sgf downloading seems to be more difficult for me than playing go