Re: Dieter's ABC of mistakes - 1
Posted: Wed Apr 14, 2021 10:45 pm
Well said lightvector. There are some 0.1 point mistakes that even 10 kyus can correctly identify. Finding small pointwise mistakes that standard human go theory has correctly identified and explained so that even weak players can get right is an interesting exercise. Sometimes it's interesting that the "wrong" answer is so little wrong that it reveals a misjudgment in the prior theory. Or it can just show the high fidelity of our theory.
Here's an example.
Focusing on 5 (the slide of 3 being about a 0.4 point loss is an example of the old theory being wrong). This extension after making the slide is a fairly common mistake which I have highlighted and explained in teaching over the years. The explanation goes like:
If you play the slide of 3, then you should continue with a closer, more connected move like the (beginner 101) joseki at c, or maybe mention b too depending on audience. This is because the big extension leaves a weakness at a. If White invades here later and you've played the slide, then you are pretty much obligated to save the f3 stone because you can't sacrifice it on a small scale by jumping into the corner. This reduces Black's good choices in how to deal with the invasion and means he can't focus so much on building the lower right area. For example this sequence is one way to deal with the invasion (when White has the ladders) and black wants to emphasise building the lower right area and central thickness, trouble is losing f3 and d2 means White is happy to get a big safe corner (actually white may still spend a move here, but this is about securing the 5 stones, the corner is already secure with c3):
Compare and contrast to the same thing happening without the slide for 3-3 exchange. Now the corner is not yet White safe territory, black could play 3-3 (a) himself to live there, so White will usually spend a move at b to secure it, and even now there's still aji at the 3-3 point to use later (e.g. a move at c could threaten some funny business in the corner). Black playing d2 to give white a stone at c3 would be terrible now, but that's what's happened above.
So by playing the slide black has made the f3 stone harder to sacrifice in a good way. Stones which are a liability: you need to save them but there's not a good way and you'd rather not, are called heavy stones. So we could say by sliding black has made f3 heavier. The reason the joseki is to extend closer with the 2 space jump is it is better connected than the larger 3 space: once you've played the slide you need to be consistent. If your want to play the big extension that's a fine move, but do so directly without making the slide exchange first. Making your stones work together in a consistent plan, and heavy stones, are 2 important topics you'll find keep cropping up. This non joseki is a good place to learn about them.
Here endeth the lesson.
That's my pre AI explanation (and I don't think AI refutes it) of what I think is correct, useful, and comprehensible theory for players even as weak as 10kyu. How much of a mistake does KataGo say 5 is? About 0.2 points. So am I going to stop teaching this because that's small and 10 kyus make 40 point mistakes in fighting and pros make 5 point ones? No. Not only is the particular shape a common one that's good to learn, the general principles of consistency and exchanges making a stone harder to sacrifice are hugely valuable. Will a 10k doing AI self review figure this out? No, human teachers are still necessary. But by not skipping this mistake they keep making as only -0.2 maybe they ask a stronger friend or post on a forum asking about it and then can get some useful human teaching.
P.S. After the slide, the 2-space HIGH extension is one which is/was rarely seen in strong play. I also learnt this as a no-no, similar to the 3-space do it without the slide if you want to do it. But it doesn't have the invasion weakness of the 3-space, so the logical argument above doesn't apply. One argument which does apply is it leaves the l3 invasions, and if you kick that at k3 then j4 ends up in a locally bad kosumi shape next to the white nobi. I seem to recall reading, probably in one of JF's books, that Go Seigen was an advocate of this high 2 space with the slide. And guess what, AI likes this move the most of all the lower side extensions! (top left 3-3 number 1 though, quelle surpise, and d7 shoulder hit also likes). Yet another example of Go Seigen's genius AI premonition.
Here's an example.
Focusing on 5 (the slide of 3 being about a 0.4 point loss is an example of the old theory being wrong). This extension after making the slide is a fairly common mistake which I have highlighted and explained in teaching over the years. The explanation goes like:
If you play the slide of 3, then you should continue with a closer, more connected move like the (beginner 101) joseki at c, or maybe mention b too depending on audience. This is because the big extension leaves a weakness at a. If White invades here later and you've played the slide, then you are pretty much obligated to save the f3 stone because you can't sacrifice it on a small scale by jumping into the corner. This reduces Black's good choices in how to deal with the invasion and means he can't focus so much on building the lower right area. For example this sequence is one way to deal with the invasion (when White has the ladders) and black wants to emphasise building the lower right area and central thickness, trouble is losing f3 and d2 means White is happy to get a big safe corner (actually white may still spend a move here, but this is about securing the 5 stones, the corner is already secure with c3):
Compare and contrast to the same thing happening without the slide for 3-3 exchange. Now the corner is not yet White safe territory, black could play 3-3 (a) himself to live there, so White will usually spend a move at b to secure it, and even now there's still aji at the 3-3 point to use later (e.g. a move at c could threaten some funny business in the corner). Black playing d2 to give white a stone at c3 would be terrible now, but that's what's happened above.
So by playing the slide black has made the f3 stone harder to sacrifice in a good way. Stones which are a liability: you need to save them but there's not a good way and you'd rather not, are called heavy stones. So we could say by sliding black has made f3 heavier. The reason the joseki is to extend closer with the 2 space jump is it is better connected than the larger 3 space: once you've played the slide you need to be consistent. If your want to play the big extension that's a fine move, but do so directly without making the slide exchange first. Making your stones work together in a consistent plan, and heavy stones, are 2 important topics you'll find keep cropping up. This non joseki is a good place to learn about them.
Here endeth the lesson.
That's my pre AI explanation (and I don't think AI refutes it) of what I think is correct, useful, and comprehensible theory for players even as weak as 10kyu. How much of a mistake does KataGo say 5 is? About 0.2 points. So am I going to stop teaching this because that's small and 10 kyus make 40 point mistakes in fighting and pros make 5 point ones? No. Not only is the particular shape a common one that's good to learn, the general principles of consistency and exchanges making a stone harder to sacrifice are hugely valuable. Will a 10k doing AI self review figure this out? No, human teachers are still necessary. But by not skipping this mistake they keep making as only -0.2 maybe they ask a stronger friend or post on a forum asking about it and then can get some useful human teaching.
P.S. After the slide, the 2-space HIGH extension is one which is/was rarely seen in strong play. I also learnt this as a no-no, similar to the 3-space do it without the slide if you want to do it. But it doesn't have the invasion weakness of the 3-space, so the logical argument above doesn't apply. One argument which does apply is it leaves the l3 invasions, and if you kick that at k3 then j4 ends up in a locally bad kosumi shape next to the white nobi. I seem to recall reading, probably in one of JF's books, that Go Seigen was an advocate of this high 2 space with the slide. And guess what, AI likes this move the most of all the lower side extensions! (top left 3-3 number 1 though, quelle surpise, and d7 shoulder hit also likes). Yet another example of Go Seigen's genius AI premonition.
My sequence shown is not of course the full or even major reason why slide and big extend is bad in AI's eyes, it's designed to illustrate, with a reasonable local continuation that could occur later, the "slide exchange makes f3 heavier" concept. I quickly checked the difference between those 2 diagrams with KG and it was about 4 points. Because the first variation diagram is so bad for black, KG suggests not even attaching on top of h3 invasion but the e4 shape building, allowing white to jump out and starting a fight. So the slide and extend is not so bad (just -0.2), but to use it well (and stop that -0.2 turning into a -2 and then a -4) black needs to abandon the standard way of using that extension and alter his play to be consistent with the slide and make all the stones work efficiently.