Pseudo sente, superficial shape, reading and other mistakes
Posted: Sat Aug 01, 2020 4:21 pm
I've recently lost 5 games in a row against 2-3k on OGS and I was unsettled by it but then read that OGS deflated their ranks. Nevertheless I have reviewed my games for mistakes and have learned a lot. My mistakes fall into these categories:
1. Pseudo sente
I play moves which intend to be sente but aren't. Sente is a big concept in what we try to learn from AI. It will often discard moves because they end in gote and there are still large moves on the board, so either it will recommend a variation that keeps sente or the big move in the first place.
2. Superficial shape
I remember from comments on a page I once compiled on SL that I treated shape too much as a static visual thing and not enough as a dynamic thing. At the time I didn't know if that was true nor how to correct that but thanks to AI reviews I'm now convinced of this bad tendency. Underneath shape lies the quest for efficiency. Good shape reaches a certain goal with a minimum amount of stones, or multiple goals with a certain amount of stones. That may result in shapes like the table shape. However, there are situations where a table shape can be made and looks like the right thing to do but it's only superficially good shape and actually inefficient.
3. Bad reading
A smaller number of mistakes were simply bad reading, where I could land a devastating blow by capturing or killing but simply missed it.
4. Other
Sometimes the move Katago suggested was something I couldn't fathom. The move I chose was much worse but not in a way I would have been able to assess in a game or independent review. They were not mistakes at my level of comprehension, rather they represent an opportunity to learn new moves.
I will post examples as exercises.
1. Pseudo sente
I play moves which intend to be sente but aren't. Sente is a big concept in what we try to learn from AI. It will often discard moves because they end in gote and there are still large moves on the board, so either it will recommend a variation that keeps sente or the big move in the first place.
2. Superficial shape
I remember from comments on a page I once compiled on SL that I treated shape too much as a static visual thing and not enough as a dynamic thing. At the time I didn't know if that was true nor how to correct that but thanks to AI reviews I'm now convinced of this bad tendency. Underneath shape lies the quest for efficiency. Good shape reaches a certain goal with a minimum amount of stones, or multiple goals with a certain amount of stones. That may result in shapes like the table shape. However, there are situations where a table shape can be made and looks like the right thing to do but it's only superficially good shape and actually inefficient.
3. Bad reading
A smaller number of mistakes were simply bad reading, where I could land a devastating blow by capturing or killing but simply missed it.
4. Other
Sometimes the move Katago suggested was something I couldn't fathom. The move I chose was much worse but not in a way I would have been able to assess in a game or independent review. They were not mistakes at my level of comprehension, rather they represent an opportunity to learn new moves.
I will post examples as exercises.
and 
are real sente although KataGo thinks this may be too good for White and recommends Black fighting back by playing at the bottom. In that case, White has severe follow up on the left side.
uses sente to play on the lower side. Black can also harass the top left but KataGo seems to prefer playing the lower side first and deal with the top left when sente can more clearly be enforced there.