Life In 19x19

Bill was making the point that Leela 11 is trained on human games from some years ago before the slide and subsequent 3-3 defence lost popularity with the AI revolution, and is thus not playing according to the theory of "these days", but those days.
As for the 3-3 the reason it's not played now isn't so much that it's slow and not because it's gote. It's generally not gote, even these days white will probably answer with the traditional low 2-space extension (or sometimes high one-space if moyo potential on the side, an old idea, or sometimes high 2-space, used to be considered a bad shape but bots seem to like it sometimes), though tenuki and then q15 attachment to make shape when black pincers is another idea depending on the whole board. The point is that exchange probably helps white more than black so black is reluctant to do it. Also black is no longer scared of white playing the 3-3 next, the bots have made us realise that white follow-up is less powerful than we thought so black doesn't need to prevent it. Why? Partly because r17 to s16 is a weak shape (imagine white r17 first and black ignored, white would not s16 as a followup but r16 to be stronger / take liberty / make more territory / help r14 more) so black q16 and o17 aren't so easy to attack (e.g. http://www.alphago-games.com/view/event ... 23/move/17), but even so black might answer r17 (at l17 or k16) and again you wonder was that really a good exchange for white this time or did black gain more in whole board development?

Pippen wrote:@Tryss: How much worse are a or b compared to what LZ suggests (because both are my favourites, and even Leela 11 like them)?

Not much worse. The 3-3 is 2% worse than LZ top choice, while Q13 is 3% worse.

You may say "it's not much", and you're right, but at this point, black has gone down from 46% to 40% in 3 moves.

Tryss wrote:You may say "it's not much", and you're right, but at this point, black has gone down from 46% to 40% in 3 moves.

Damn it. We need new weight files for LZ that doesn't depress me so much.

Bill Spight wrote:

Aidoneus wrote:

Pippen wrote:@Tryss: How much worse are a or b compared to what LZ suggests (because both are my favourites, and even Leela 11 like them)?

Isn't a considered too slow and gote these days?

Leela 11 was trained on human play.

Aidoneus wrote:Sorry Bill, I don't follow you here. I was refering to the 3-3 defense of the corner, which seems gote and slow...but then I am still just a humble little kyu.

What Uberdude said,

Also, the 3-3 response to the slide started being omitted in the 1980s, it was much later that it came into disfavor. That made sense to me, because I had always questioned the slide. And the 3-3 follow-up (when the response is omitted) makes the slide look inefficient, eh? But when Leela 11 was trained, the 3-3 response was still popular enough for Leela to like it.

Additionally, I think I read in some thread here about comparative efficiency of adding a stone to different areas based on the local balance of power (number of stones per side), which I thought might apply in this position. Thank you in advance for any further insight!

Well, yeah, the 3-3 response puts Black one stone ahead locally, and threatens White's potential base. But, Leela 11 aside, the bots don't like it, and they don't particularly care for the slide, either.

Bill Spight wrote:

Aidoneus wrote:Additionally, I think I read in some thread here about comparative efficiency of adding a stone to different areas based on the local balance of power (number of stones per side), which I thought might apply in this position. Thank you in advance for any further insight!

Well, yeah, the 3-3 response puts Black one stone ahead locally, and threatens White's potential base. But, Leela 11 aside, the bots don't like it, and they don't particularly care for the slide, either.

But I was thinking that adding an approach stone to one of the other corners was more efficient per something that I read...I think in a thread here but I cannot find it at the moment. Something about getting one vs one or two versus one being more important? I guess I see it as percentage change in relative strength, though it wasn't expressed that way. Does this sound familiar? Maybe I was dreaming?!

Aidoneus wrote:

Bill Spight wrote:

Aidoneus wrote:Additionally, I think I read in some thread here about comparative efficiency of adding a stone to different areas based on the local balance of power (number of stones per side), which I thought might apply in this position. Thank you in advance for any further insight!

Well, yeah, the 3-3 response puts Black one stone ahead locally, and threatens White's potential base. But, Leela 11 aside, the bots don't like it, and they don't particularly care for the slide, either.

But I was thinking that adding an approach stone to one of the other corners was more efficient per something that I read...I think in a thread here but I cannot find it at the moment. Something about getting one vs one or two versus one being more important? I guess I see it as percentage change in relative strength, though it wasn't expressed that way. Does this sound familiar? Maybe I was dreaming?!

No, you weren't dreaming. In general, with non-random placement of stones, and unless there is a capture or threat to capture, the more stones in a region of the board the lower its local temperature.

For instance, around three centuries ago in Japan it was popular to start games with a 3-4, then a 5-3 approach, then a pincer, then play in an open corner, often with the same pattern. Over time, players learned through experience that it was better to play in an open corner than to play a pincer. Later they started playing in an open corner instead of playing the approach.

OC, both local and whole board considerations can override this heuristic. For instance, bots usually prefer a corner approach or three-three invasion to playing in an empty side.