There's been quite a lot of activity recently on the thread
Reviewing SL using LZ and criticism. I am making this thread as both an interesting study of Go positions, but also highlights that applying the standard of "approved by LZ" is actually an incredibly high standard of quality that much of Go literature fails to meet, whether random Western amateur's scribblings on SL or the thoughts of the great honorary Kisei Fujisawa Shuko 9p, widely admired for his skill in the opening and early middlegame, in a published book. (This is not to say I don't think we should strive to improve them with the tools we now have).
The problem(s) I will study come from the book "Shuko: the only move, volume 2, fighting middlegame collection". These come from his comments on study session games of 'young' players so I believe do really reflect his thoughts (via translation) rather than those of a ghost writer, as can happen with other books with a pro's name on the cover. I did some similar anaylses a while back on a thread in the computer go subforum.
For the first analysis, I used Ah Q Go running a 15-block LZ on my mobile phone with a few hundred to few thousand playouts so that we can see how strong a phone is (does it have a better intuition than Shuko but make reading mistakes?), and then I will do a second-pass using many playouts of LZ/Elf/KataGo on my GPU.
Quote:
The extension of
starts an intense skirmish. Is it possible to kill two birds with one stone by pressing White closely, while reinforcing Black's own position? It must be a flexible move, that is a hint.
- Click Here To Show Diagram Code
[go]$$Bc Problem 11, Black to play
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O X X X . . Q . . O . X . . . . . |
$$ | . O . O . . O . . , . . . . . , X . . |
$$ | . . . O X X . . . . . . . . . . . . . |
$$ | . . . . O . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . X . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O O . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . O . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . O . . . |
$$ | . . X X . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
Before giving his answer, Shuko gives a brief review of any major mistakes he sees so far. How well will this agree with LZ? Without even checking order of play I can predict that it won't like:
- white's kosumi at lower left, that's a slack shape, should attach
- lower right hasn't been 3-3ed or approached, that will have been preferred as a bigger and more urgent move over other moves already played
- top left looks like that table shape joseki, but white didn't push through like AG taught us (but the pincer is looser than normal, will that change things?)
- black one point jump on left side is rather vague and not doing much efficiently
- white hanging connection top left should probably have been descend.
Here is the course of play, white was Yokota Shigeaki 9p, black Kenmochi Jo 7p. Shuko doesn't criticise anything here just saying how black would have made the mini-Chnese if 5 responded at 9 and "When white starts operations in the upper left with 6 and 8 black first fixes the shape in the lower left with 9 through 13, and then responds with 15, setting a rapid pace."
- Click Here To Show Diagram Code
[go]$$Bc
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 3 . . . . . . . . . . . . . . . |
$$ | . . . , . . 8 . . , . . . . . , 1 . . |
$$ | . . . 6 . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 7 . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 5 . . . . . . . . . . . . . . . . |
$$ | . . . . 0 . . . . . . . . . . . . . . |
$$ | . . . 2 . . . . . , . . . . . 4 . . . |
$$ | . . . . . 9 . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
Phone LZ likes 6 (or other approaches) over back off at 9 (-2%, as this is weaker LZ than recent 40-blocks the %s are smaller), which is unsurprising given
The opening gospel according to LZ. 7 is of course bad (-3%), being an even looser pincer than the ones humans commonly play which bots still don't like, attach under at c15, or shimari top left or approaching/invade 4-4s are better moves, as the gospel tells. (NB I think that amateur dan players like myself and maybe even kyus can play a better first 20 moves of the game than these pros simply by following that gospel). 8 is also bad, should attach c17 (corners are king) or approach top left. 9 is ok, but would prefer to c15 attach first to take corner points and eyes in sente. 10 kosumi is bad (-5%, worse move of game so far) as predicted.
No comments from Shuko on this
- Click Here To Show Diagram Code
[go]$$Bcm11
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 0 X . 9 . . . . . . . . . . . . . |
$$ | . . . 8 . . O . . , . . . . . , X . . |
$$ | . . . O 5 7 . . . . . . . . . . . . . |
$$ | . . . . 6 . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 4 2 . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . O . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . O . . . |
$$ | . . 1 3 . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
Finally with 11 we get a #1 LZ move, 12 again off-topic (c17 or top right), 13 is correct, these pros can at least answer sente moves correctly. For 14 LZ wants to play an interesting move that was unexpected to me at d9: this prepares to block at c4 to develop the left side and puts some pressure on the pincer, c7 is -7% as (my interpretation) when black plays c5 you got no territory, no eyes, have a cut at e7 and aren't attacking d9, you just made a heavy little wall. Attaching at 15 to cut big knight is a move AG didn't like either, LZ recommends f14 3% better. Instead of 19 table shape wants e17 extend (1k playouts) but when played actually prefers table a bit, and then doesn't want to push through but hane in corner like the pro.
Shuko no comments til 24, see later
- Click Here To Show Diagram Code
[go]$$Bcm21
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O X 1 X . . 6 . . 4 . 5 . . . . . |
$$ | . 2 . O . . O . . , . . . . . , X . . |
$$ | . . . O X X . . . . . . . . . . . . . |
$$ | . . . . O . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . 3 . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O O . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . O . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . O . . . |
$$ | . . X X . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
LZ says c18 descent considerably (6%) better as I predicted, I suppose the pro didn't like that black could then cut at d14 whereas hanging connect allow that to be captured with d13, but cut is a bad move and w simply ataris from under and pushes and then takes sente happy with a large corner and still cutting stone aji. In fact after c18 best move for black is not cut but c13 peep. The weird 23 is only -3%. So LZ, and also me following the gospel, see numerous mistakes so far which Shuko didn't see as major enough to comment on, how will they compare in size to 24 which he does criticise?
He says
Quote:
I suppose that the result through black 23 is reasonable, but what about white 24 here? I guess that white thinks if the group on the upper side is settled then the black group in the upper left can be attacked. Instead, one wants to attack the white corner with
[below]. Should black defend with 2 through 6 the extension of 7 works effectively for white.
In regards to using
to make a pincer on the upper side,,, well, black is weak in the upper left but there is no pincer that fits the bill precisely.
I suppose that for
a is also possible. If white can slowly get into the centre, black will become weak on the left and upper sides.
- Click Here To Show Diagram Code
[go]$$Wc
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O X X X . . 7 . . . . 3 2 4 . . . |
$$ | . O . O . . O . . , . . 5 . 1 , X . . |
$$ | . . . O X X . . . . . . . . . . . . . |
$$ | . . . . O . . . . . . . . . . . 6 . . |
$$ | . . . . . . a . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . X . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O O . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . O . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . O . . . |
$$ | . . X X . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
LZ agrees that high approach is better than the wedge, no surprise there, corners before sides as the proverb and gospel tell us. But it's not by much, following wedge it agrees with 25 and 26 and white is at 52%, but after high approach it's only 54%, ie this mistake is several times smaller than other mistakes so far which went unremarked. Interestingly LZ thinks the best response to high approach is the unusual 2-space low pincer than AG also recommended in teaching tool. And it thinks low approach was 1.5% better than high, so Shuko's suggested move is almost as much a mistake from the best move as the real game's move was from Shuko's suggestion. What of Shuko's judgement that if black follows the standard joseki white 7 is good? LZ strongly agrees, white is up to 62% now, but that's because black played bad moves to make Shuko's move look better than it deserved (a common trait I've noted in his sequences). 2 was a little bad (and Shuko did at least consider a pincer but say nothing looks satisfactory, which seems fair enough if the best move is the wacky pincer), but the chief fault lies with the cooperative black 6. As bots have taught us the hanging connection of 5 is not sente so black should have taken the key point of mutual extension and eyespace point of j17 for himself for 51%, instead of losing 13% at r14. So Shuko's illustrative sequence contains a mistake from the opponent
far larger than any in the game so far and the initial mistake he was correcting. It was at least given as a conditional "Should black play 2 etc..." and he did consider pincer instead of 2, but he didn't, or at least not recorded in the book, consider a stronger resistance for black further down his sequence. I could well imagine if black did play such an extension he would not be too peturbed and press the corner as those 2 areas are generally consider miai in the joseki, but LZ is saying that relying on this joseki judgement that they are miai is wrong here because the top side is more valuable due to black's weak group there. What white should have done is either solid connect because that is more sente than hanging, but even so black should STILL tenuki to j17 and allow white q14 or r15, or just directly take the vital point at j17 and allow black to cut, or if black defends at f14 then white can o16 connect.
I wouldn't claim that Shuko should have highlighted the move with the absolute biggest win% drop in the game so far as the one for criticism, as the move with the most valuable teaching lesson that can be well extracted into explainable ideas and re-applicable valuable principles might not be the same, but to highlight a 2% mistake (vs his suggestion) and then make a 13% mistake in his sequence to justify why his move is better, whilst ignoring several >5% mistakes in the game means I give a pretty poor "LZ approved" score so far.
Login
Register
FAQ
Search
Uberdude
LZ gave black 50% (and after f14 turn too as best move). The real game L15 took that down to 41%. After Shuko's j15 LZ gives black 46% on low playouts when it has white answering at c4, but this drops down to 43% as it finds an exquisite riposte with a few hundred playouts. First what of Shuko's sequence, as before it is given with some conditionallity on white's replies "would be standard", "if white plays 4". Well, having an idea of what to do if standard is all well and good as a starting point for playing at a high level, but really you should be prepared for your opponent to resist your plans and try to find an even better move. Phone LZ does consider m15 jump and gives black 47%, but other moves it likes more (but less than the amazing #1 for later) are p17, q17 and q16 operations against top right shimari, l15 different jump, e12 peep, c4 block, h14 or g13 cut. All these moves make sense to me as candidate moves a strong player could consider. If white indulges Shuko with m15, then LZ thinks 3 is also soft, it prefers to peep at L16, white connects and then k18, avoiding this peep is probably why it had a slight preference for l15 jump instead. It also likes tenuki to b5, which makes sense as white c4 was a good move before so it's big to connect up there. If m15 was soft before (-5%) then m13 jump is even softer (-6%). White should instead begin operations to cut j15 with e15, h14 or other moves in that area. LZ is not a fan of 5 either (-6%), preferring the thick f14 turn (with maybe l16 peep first), which does seem much more Shuko thick style to me.
It's not like humans did not know that the hanging connection is not sente. IIRC, Fujisawa himself, in a game against Sakata, did not treat the hanging connection as sente but played the equivalent of K-16 in this orientation. (OC, the rest of the board was completely different.) Everybody needs a good editor.