Top 10 Shocking Winning Percentages of the AlphaGo Teaching Tool, Part 2

Continuation (duh) of WeiqiTV‘s top 10 selection, as cited from “AlphaGo教学工具之十大恐怖胜率 下集”. Keep in mind that the percentage values always show Black’s winning probability. Black goes for big numbers and White goes for small ones.



Good old large avalanche.


Old variation, bad for Black. Forget this sequence.


Modern joseki, until recently considered mostly even.

The illusion of this joseki’s equilibrium is shattered by the following combo that basically punishes Black’s shape.


This sequence creates a damezumari in Black’s shape and leaves him with only ~36%. What went wrong?

Apparently, the choice of avalanche is a mistake:


However, it is immediately evened out by White’s failure to find an obscure variation of the small avalanche:


I don’t get the tenuki at the end of the sequence, check Tool.

This implies that Black made more mistakes further down the former joseki path:


Black should descend without turning first in the corner. This has been considered a rookie mistake ever since this joseki existed. Alas, I memorised all those variations in vain.


This would happen and, however White captures the two stones on the left, Black can find a way to sacrifice them efficiently.

The former joseki contains another mistake:


AlphaGo does not even consider turning at b12 at this point :S Nobi is the only move.


Black only turns at 12 after White has already played inside at 9. Though in this case it’s just an exchange to sacrifice better. Never was AlphaGo planning on pulling out those stones.



Lee Changho’s sure-win fuseki from way back.

Frankly, I’m kinda surprised that this one made it to the top 10. I thought it common knowledge that White got ripped off in this sequence. (Black ends in sente, and the Teaching Tool suggests the lower left 3-3 next. Though any move on the upper side is probably still good.)

By rearranging the order of moves, it becomes obvious that White got the short end of the stick.

This slideshow requires JavaScript.

Tewari narration: In the starting position of White’s two-space extension, Black makes an atrocious attachment 1 that turns out to intiate some amazing exchanges due to White’s overly timid responses 4, and later 8. Then Black even gets to peep at 9 in sente. Black’s only minus from throwing in at 11 is negligible.

Thus it is a long-known fact that White needs to avoid the initial sequence at all costs.
Btw, peeping at 19 immediately is necessary, lest White harrass Black’s shimari right away. (Check the Tool.)

AlphaGo suggests White do the following:


This leads to a sequence way beyond my understanding:


After a couple mysterious exchanges, the tree splits up with A or B for White next.

White could also…


…turn and jump out up to 10.


The most dramatic mistake is the last nobi 18 of the initial sequence. It loses another ~13%:




This one is mostly irrelevant to me because I play neither 5 nor 11 anymore. It doesn’t seem to be a good opening for Black, unless White makes the horrible mistake of connecting under with 20 and 22.


There go 12.5% in one move, but that’s not all.


If White insists on connecting next move, (which White will,) Black’s 40.9% instantly turn into 59.6%.

The alternative would look like this:




23 at 9, 27 at 18, 29 at 16.

Almost every move Black played in this joseki is bad 😮 In the end Black is left with a measly 22.8%.


I’m not going to point out every single move (because this is a summary). Screenshotting is a lot of work. Please click through the Teaching Tool.

Apparently, the largest drop happened at this point:




At last, #1) banishes another widespread sequence to the joseki graveyard.

I assume that everyone knows the joseki, and the ladder variation when White plays 5 at A.


If you don’t, shame on you, but it is of no more importance anyway.


Because if White plays the formerly forbidden sequence, Black is already losing.


WeiqiTV neglected to point out where Black went wrong. It is perhaps the first pincer 17 that leads to Black’s inevitable disadvantage. The attachment 19 also seems to be a bad move if it is inferior to even a move like that second-line kosumi proposed by AlphaGo.


In a different position, the joseki attachment is again criticised:


Continuing locally seems to put Black at a disadvantage. But if Black insists on doing so, we’ll find a recurring sequence in the Teaching Tool that I’m sure no one has ever played before:


Hope this helps.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s