It is currently Tue Feb 18, 2020 8:15 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 113 posts ]  Go to page Previous  1, 2, 3, 4, 5, 6  Next
Author Message
Online
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #81 Posted: Wed Dec 11, 2019 6:31 am 
Honinbo

Posts: 9417
Liked others: 2945
Was liked: 3150
xela wrote:
Applebaps wrote:
Also, a lot of "outdated" techniques still work just fine against human opponents. I know from firsthand experience that psychologically-motivated moves/styles often render an opponent weaker than they otherwise would be. It won't work on a bot, but I'm not playing bots, so who cares?

Now here's where we depart ways. Yes, some club players do things that they know aren't quite right, but it works against amateur opponents. And if you fight fire with fire, you might be able to match the other players in your club. But you'll never surpass them until you learn to stop setting fire to things!


Well, Applebaps, is your goal to improve your game, or to beat weak players? If the latter, then making plays that they do not understand can help you to do so. If the former, better to pretend that they are good players and still try to beat them. :)

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #82 Posted: Wed Dec 11, 2019 6:48 am 
Gosei
User avatar

Posts: 1516
Location: Ghent, Belgium
Liked others: 218
Was liked: 681
Rank: KGS 1d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Uberdude wrote:
I think John's point about people enjoying Go as fans rather than (or as well as) players answers ez4u's point about the difficulty of how to apply LZ's lessons when you are playing a game yourself. You sinply don't need to! Exploring pro games, problems, exploring variations etc with LZ, such as I did here on book by Fujisawa Shuko, is an interesting and enjoyable activity for me (and I hope other readers too) without requiring validation that I can play as well as LZ here in my own games. Just look at that f12 attachment, it's so cool! And by exploring the follow ups I can see its purpose and followups depending how the opponent answers. If I can imagine such moves in my own games then that's a bonus, but it's already a positive without me playing.


This sums it up better than I could myself. I enjoy studying Go and I find it impossible to ignore the recent evolution due to AI. I was already quite inept at applying conventional wisdom to my game and that won't change much with AI generated ideas. I find these AI generated ideas not that much harder to understand than conventional wisdom. They are intriguing, for sure, unsettling, often, counterintuitive sometimes, but intuitive in nature, because of their probabilistic essence. They are not super weird moves which only work because of a tesuji 80 moves down the line. Uberdude's gospel of AI openings is a neat, well articulated example of what we can learn.


This post by Knotwilg was liked by: Applebaps
Top
 Profile  
 
Online
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #83 Posted: Wed Dec 11, 2019 7:05 am 
Honinbo

Posts: 9417
Liked others: 2945
Was liked: 3150
ez4u wrote:
I think I am more on JF's side. What is truth? You are part way through a game. You have the opportunity to ask for LZ's choice for the next move. However, you are not allowed to see any of the variations and will not have any further recourse to LZ throughout the rest of the game. LZ returns a move. Should you play this move or an alternative that you choose based on the heuristics that you have previously studied and utilized up to now?

If you play LZ's recommendation, how do you plan to continue to choose your future moves in the game?


I think that there is a parallel, here, with joseki. Do you play joseki or not? The saying goes, Learn joseki, lose two stones. I took that to heart, and did not learn joseki, except by picking them up, until I was a dan player. Before then, I did read a book about joseki mistakes, however. Playing by rote is not good. As John Fairbairn says, go is a mind sport. And, as Znosko-Borovsky says,
Znosko-Borovsky wrote:
Do not make the opening moves automatically and without reflection.
Do not seek to memorise variations, try to understand them.


So my answer is, neither. I would not just play the LZ move, nor would I play the presumably inferior move that first occurred to me. (Let's not bias the choice by letting me invoke thought for one play but not the other.) Instead, I would apply reflection and try to understand LZ's move. Then I would make my choice.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins


This post by Bill Spight was liked by: Applebaps
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #84 Posted: Wed Dec 11, 2019 7:42 am 
Oza

Posts: 2461
Liked others: 15
Was liked: 3547
I find the argument about being shown a new move by a bot as a reason to value bots spurious.

Yes, it's interesting, if you like that kind of thing. Yes, it probably is a better move.

But you get that kind of insight over and over again in playing over human pro games. With the humans you also, in many cases, get a useful human commentary as to why it is a good move.

In fact, over the years the one thing that has improved my game much more than anything else is playing over a pro game and seeing a move to which my reaction is: "Oh, I didn't know you could do that!"

I would then try that move in my own games, no doubt with a varying degree of understanding and success - but on the whole I felt I was understanding more and getting better results. Which is another way of recommending Bill's advice to try new moves. Yes, but note - you can do that with human moves, too. And, again, you get a human/pro commentary, not a bot/amateur commentary.

In my view of things, when we have a case where it can be shown that a pro recommendation is wrong according to a bot (as I have done myself a few times here), my reaction is not to worship the bot. I prefer to try to see if I can modify my understanding of the traditional wisdom so as to retain a grasp on the game. E.g. a valid question may be: is the pro's advice generally right but not in this particular case - and if so what is particular about this case? That seems likelier to lead to quicker improvement. That in fact, from what I read, is what the Japanese pros have done when they started speaking not just of overconcentration but bots making extra-early overconcentration.

Moderation in all things. It's long lane that has no loaf on the bread.


This post by John Fairbairn was liked by: Applebaps
Top
 Profile  
 
Online
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #85 Posted: Wed Dec 11, 2019 7:44 am 
Honinbo

Posts: 9417
Liked others: 2945
Was liked: 3150
John Fairbairn wrote:
But I think the real topic has revealed itself, here and in other threads, as 'old go wisdom is old wives' tales'.


I beg to differ. Some of it is, to be sure. Already some joseki have become obsolete. They would probably have become obsolete in a century or two without AI, but they are now. So what? Is that a bad thing?

Quote:
The first thing I don't properly understand is the very strong urge to debunk all old wives. In real life, those old wives were your mothers and grandmothers, and they did a pretty good job of getting you where you are today. Much of the time they used their intuition and not their consciousness, made plenty of mistakes, too. But, if you could live your life again, would you still like to be brought up at home, or would you prefer to be brought up in a germ-free isolation ward of a hospital?


I don't think that's a good analogy. I was brought up by parents who did not believe in caning their children, and I am glad of it. Several of my schoolmates were disciplined with belts and yardsticks, so that was still part of the culture.

By analogy with learning at mother's knee, Takagawa was my go mom. Now, thanks to AI, I have found out that some of what I learned from Takagawa, not all, must be discarded, particularly with regard to plays on the side. Plays which a few years ago were obvious, now are to be avoided. I have to learn new plays. So what? Old dogs can learn new tricks. Or die trying, anyway. :lol:

One thing the bots tell us is of little importance, at least as commonly understood, is direction of play. Amazing! Heresy! But what can you say? I certainly cannot base game reviews on it, anymore. Or recommend Kajiwara's book about it, even though he is entertaining and thought provoking. Phlogiston, anyone? Geocentric cosmology? Direction of play may become important late in the opening or in the middle game, but I don't know that yet.

The saying is that science marches on, funeral by funeral. Well, not over my dead body! At least not yet. ;)

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins


This post by Bill Spight was liked by 3 people: Applebaps, dfan, Gomoto
Top
 Profile  
 
Online
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #86 Posted: Wed Dec 11, 2019 8:16 am 
Honinbo

Posts: 9417
Liked others: 2945
Was liked: 3150
Addendum to my response to xela about the solid connection below, which he learned from 38 Basic Joseki.

Click Here To Show Diagram Code
[go]$$Bc
$$ . . . . . . |
$$ . . . . . . |
$$ . . O X . . |
$$ . . O X . . |
$$ . O X 1 . . |
$$ . . . . . . |
$$ . . . . . . |
$$-------------+[/go]


I replied that the bots teach the sagari, instead.

Click Here To Show Diagram Code
[go]$$Bc
$$ . . . . . . |
$$ . . . . . . |
$$ . . O X . . |
$$ . . O X . . |
$$ . O X . . . |
$$ . . 1 . . . |
$$ . . . . . . |
$$-------------+[/go]


In that case, the following situation could arise.

Click Here To Show Diagram Code
[go]$$Bc
$$ . . . . . . |
$$ . . . . . . |
$$ . . . . 4 . |
$$ . . . 2 3 . |
$$ . . O X . . |
$$ . . O X . . |
$$ . O X . . . |
$$ . . 1 . . . |
$$ . . . . . . |
$$-------------+[/go]


Black is then faced with the following question, which she would not face after the solid connection for :b1:.

Click Here To Show Diagram Code
[go]$$Bcm5
$$ . . . . . . |
$$ . . . . 3 . |
$$ . . . 1 O 5 |
$$ . . 2 O X . |
$$ . . O X 4 9 |
$$ . . O X 7 . |
$$ . O X 6 8 . |
$$ . 0 X . . . |
$$ . . . . . . |
$$-------------+[/go]


Is this good for Black, or not? Good question. :) To quote my late wife:

Winona Adkins wrote:
At some point, doesn't thinking have to go on?

:D

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins


This post by Bill Spight was liked by: Applebaps
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #87 Posted: Wed Dec 11, 2019 9:40 am 
Gosei

Posts: 1479
Location: Earth
Liked others: 555
Was liked: 236
We are not only watching a very interesting discussion about the old and new ways in go with very good arguments on both sides. We are also watching the eternal struggle between chaos and order near the battleline. Therefore we will not solve the question at hand if we choose one side or the other, or if we think of the question as only be related to the domain of go.

It is good advice to honour our predecessors and look for the new ways at the same time. And even if the only reason would be: Just to enjoy go even more.


This post by Gomoto was liked by: Applebaps
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #88 Posted: Wed Dec 11, 2019 9:56 am 
Judan

Posts: 6323
Location: Cambridge, UK
Liked others: 364
Was liked: 3434
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
ez4u wrote:
I think I am more on JF's side. What is truth? You are part way through a game. You have the opportunity to ask for LZ's choice for the next move. However, you are not allowed to see any of the variations and will not have any further recourse to LZ throughout the rest of the game. LZ returns a move. Should you play this move or an alternative that you choose based on the heuristics that you have previously studied and utilized up to now?

If you play LZ's recommendation, how do you plan to continue to choose your future moves in the game?

Although I think this was directed at Knotwilg, I think my previous example explaining my thinking in a real game at https://lifein19x19.com/viewtopic.php?p=251392#p251392 is relevant. I didn't ask LZ on my phone what to play for 19, but I had looked at 2-space high pincers to 4-4s a bit with LZ as AK loves that move so I expected some to happen. From playing around with variations with that and other pincers, I knew that LZ likes to activate the c14 stone pretty early when white makes the old fashioned c15 connection rather than b14 crawl (e.g. this LZ game I played on Fox a year ago https://lifein19x19.com/viewtopic.php?p=238695#p238695); though b14 is unsatisfactory with 2-space high so is presumably why bots don't like that pincer but e.g. 1-space low or 3-space high are played. So I didn't know that LZ would approve of 19 as e14 jump, but I had a pretty good idea it would. Of course I hadn't studied this exact board position with the 2-space enclosure and white wedge on lower side, but I do now know that bots are happier to tenuki an approach to 4-4 that we used to think so 19 as tenuki was more worth considering. Using the e14 jump to make c9 more severe is my own thinking and planning and LZ generally approves of how I played so I think this was a successful synthesis of LZ and my own ideas. That I then made a mistake 20 moves later and spoilt the lead that I had built up doesn't invalidate that. I think LZ makes my best moves better, but doesn't much improve the level of my worst moves, and that tends to be what decides who wins games.

ez4u wrote:
I have great difficulty in using the ideas from AI games. It is easy to sit passively and watch LZ generate variations and moves in Lizzie. It is a very different challenge to take those ideas and turn them into consistent and successful play in real life.

Another position from that game where I played a good move which I'm sure I wouldn't have a few years ago in its own thread: https://lifein19x19.com/viewtopic.php?f=15&t=17130. What made me think of this move? I wouldn't say it is a totally AI style move, pros have played this sort of one-point jump side reduction before AlphaGo (the dancing elephants of Lee Sedol and Mok Jinseok), and maybe I would have played it in the alternate universe of no AG and my spending the same amount of time replaying pro games as I do in this universe, but I think AI existing, both directly as I use myself and how it has influenced pro play increased the chances of me finding that move.


This post by Uberdude was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #89 Posted: Wed Dec 11, 2019 10:11 am 
Dies in gote
User avatar

Posts: 64
Liked others: 129
Was liked: 24
Rank: DDK Life
Universal go server handle: Applebaps
Online playing schedule: Evenings M-F UTC-9, Saturday mornings
Too many good posts to quote them all! I'm enjoying this discussion, as it gets right to the root of why we play this game we (presumably) love. I think, on some level, the AI revolution has forced us (the last bastion of humanity's superiority to AI, in a way) to finally question ourselves in the way that Chess and other games have had to go through already many times. If no human can ever touch a computer at this game, does that mean we should continue, or not? What's the benefit we get? What's the point of it, ultimately?

Reactions vary widely, and I don't think there's a right answer, only how we feel and how we act on those feelings. In any case, I hope my intent comes across in my post that I'm not trying to yell at anyone or say that anyone posting in here is objectively wrong or anything. I don't think that at all.

RE: players and fans - I'm both a player and a fan of Go. I don't think the split is so clear, personally.

RE: the complications invited by AI play - this is the crux of why I tend to look askance at AI's advice, to be honest. At a fundamental biological level, our brains are built "to keep us alive", not "to play Go". AI's brains are built from the ground up "to play Go." Chess had this same problem, where computers would happily enter into "sharp" or complicated positions from which extensive calculations were required but which offered a lot of tactical flexibility. Our brains don't have nearly that level of calculation ability. We're never going to be able to emulate them. So why not just be true to ourselves? I'd rather play moves I understand, I'd rather play like myself, even if it means I get cut apart on the board. I want to play honestly. Here we get into a larger existential question, like, "what does it mean to play like myself, who am I exactly" haha. And I believe the self is both illusory and ever-changing. Who knows, maybe someday we'll see humans who can teach all these greedy bots not to turn the whole board into dame, by outclassing them and making them be more careful. But that day won't be in my lifetime, I don't think.

RE: obsoletion of joseki - It's true that some joseki are considered "obsolete", but others have also been revived (like attach and extend)! Can we really say any joseki is obsolete, then? What if AI in 10 years says it's fine now? Will we just dance around at their whims, always staunchly convinced of our own (their own) rightness from moment to moment? In any case, moves can be said to be good or bad for one player, but like, the board is still almost always playable. The finer gradations of the quality of moves by 1 or 2 percent are an interesting rabbit hole, but imo they're a rabbit hole nonetheless. I think we ought to play moves that feel natural and good, and see what happens.

Reducing the game of Go, a game with more possible board positions than there are visible protons in the known universe, to an equation to be solved... it should be a crime, haha. This is a beautiful, near-infinite game that ought to be approached with more respect and even a bit of awe. When we start just throwing down stones with numbers in mind rather than the art, I think we lose something essential to the experience. And I say this as someone with a bachelor's in mathematics, by the way. We don't HAVE to go down this road just because the tech exists.

_________________
Joseki (n): 1. Japanese term meaning "when Jo lives in seki."


This post by Applebaps was liked by: Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #90 Posted: Wed Dec 11, 2019 10:30 am 
Dies in gote
User avatar

Posts: 64
Liked others: 129
Was liked: 24
Rank: DDK Life
Universal go server handle: Applebaps
Online playing schedule: Evenings M-F UTC-9, Saturday mornings
Bill Spight wrote:
xela wrote:
Applebaps wrote:
Also, a lot of "outdated" techniques still work just fine against human opponents. I know from firsthand experience that psychologically-motivated moves/styles often render an opponent weaker than they otherwise would be. It won't work on a bot, but I'm not playing bots, so who cares?

Now here's where we depart ways. Yes, some club players do things that they know aren't quite right, but it works against amateur opponents. And if you fight fire with fire, you might be able to match the other players in your club. But you'll never surpass them until you learn to stop setting fire to things!


Well, Applebaps, is your goal to improve your game, or to beat weak players? If the latter, then making plays that they do not understand can help you to do so. If the former, better to pretend that they are good players and still try to beat them. :)


I would like to directly answer this and say that my goal is (partly) to improve my game and definitely not just to beat weak players. When I say a move would work on a human opponent that wouldn't work on a bot, I'm not talking about anything tricky or strange. I just mean that people aren't perfect calculating machines and you have a lot more leeway in your games against them than you do against something built specifically to tear you apart on a Go board.

As a side note, you can also chat with them, hang out with them, and there's great value to be had around and off the board from knowing and playing human Go players! As we all know from posting here :D

Basically my goal is threefold:
1. become stronger at Go
2. enjoy myself while playing, and help others enjoy themselves and get stronger
3. be a human being in the world, not just "a Go player"

_________________
Joseki (n): 1. Japanese term meaning "when Jo lives in seki."

Top
 Profile  
 
Online
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #91 Posted: Wed Dec 11, 2019 11:20 am 
Honinbo

Posts: 9417
Liked others: 2945
Was liked: 3150
Applebaps wrote:
RE: the complications invited by AI play - this is the crux of why I tend to look askance at AI's advice, to be honest. At a fundamental biological level, our brains are built "to keep us alive", not "to play Go". AI's brains are built from the ground up "to play Go." Chess had this same problem, where computers would happily enter into "sharp" or complicated positions from which extensive calculations were required but which offered a lot of tactical flexibility.


It sounds like you may be confusing top go bots with stockfish-like chess engines, which rely upon tactical superiority. Even today's top go bots are inferior to even dan level humans in local deep search. Their superiority relies mainly upon "intuition" and global strategy.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #92 Posted: Wed Dec 11, 2019 11:23 am 
Dies in gote
User avatar

Posts: 64
Liked others: 129
Was liked: 24
Rank: DDK Life
Universal go server handle: Applebaps
Online playing schedule: Evenings M-F UTC-9, Saturday mornings
Bill Spight wrote:
Applebaps wrote:
RE: the complications invited by AI play - this is the crux of why I tend to look askance at AI's advice, to be honest. At a fundamental biological level, our brains are built "to keep us alive", not "to play Go". AI's brains are built from the ground up "to play Go." Chess had this same problem, where computers would happily enter into "sharp" or complicated positions from which extensive calculations were required but which offered a lot of tactical flexibility.


It sounds like you may be confusing top go bots with stockfish-like chess engines, which rely upon tactical superiority. Even today's top go bots are inferior to even dan level humans in local deep search. Their superiority relies mainly upon "intuition" and global strategy.


No, I'm thinking specifically of Leela Zero and AlphaGo Zero here. They do have global strategy but they also love to start fights and mess up the board, treating influence as if it means nothing, moyos as if they're already reduced, and they cut cut cut all day knowing they'll win the ensuing fight.

I've played a LOT against these engines. I'm not dismissing them out of ignorance.

_________________
Joseki (n): 1. Japanese term meaning "when Jo lives in seki."


This post by Applebaps was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #93 Posted: Wed Dec 11, 2019 11:36 am 
Oza

Posts: 2461
Liked others: 15
Was liked: 3547
I think this discussion has reached a point where we start going round in circles. At least I feel it's time for me to leave it.

FWIW what I go away with is a feeling that the debate can be summed up as follows:

1. Side A: We can learn from AI but it will take a long time. Side B: Agreed.

2. Side B: We can learn from humans and can learn NOW. Sheldonian Side A: Disagree! How can a bunch of frightened dinosaurs even begin to think not using science can possibly work? Their approach has only got them to 9-dan level. Almost as bad as being an engineer from MIT! And that's my spot.

My vote goes to Penny.


This post by John Fairbairn was liked by: Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #94 Posted: Wed Dec 11, 2019 11:43 am 
Judan

Posts: 6323
Location: Cambridge, UK
Liked others: 364
Was liked: 3434
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Applebaps wrote:
No, I'm thinking specifically of Leela Zero and AlphaGo Zero here. ...
I've played a LOT against these engines.

You've played against AlphaGo Zero? Lucky you!

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #95 Posted: Wed Dec 11, 2019 12:55 pm 
Dies with sente

Posts: 117
Liked others: 1
Was liked: 15
Rank: AGA 4 dan
Universal go server handle: telegraphgo
John Fairbairn wrote:
I think this discussion has reached a point where we start going round in circles.


Agreed - since it feels like people have founds the positions they feel comfortable with, so should we now actually make a decision on what to do about old SL pages? Perhaps we should start a poll, or series of polls, with a bunch of options ranging from the extremes of rewriting with primarily AI analysis to leaving them 100% alone?

I find it fascinating and highly peculiar that anyone would choose not to consult Go Seigen's opinion. Gossip can be found everywhere and anywhere, Go Seigen's opinion cannot. It feels like opting not to take a free shot on goal - sure, the goalie might block it, but what if they don't? It's likely none of us would understand what he had to say, but what if some small part of us did, just a little bit more than before? I'd pay money per move to show him specific positions and hear him say no more than "L8". Luckily, I don't have to pay, because Leela is free! A few robotic tendencies are easy enough to ignore, when I can actually teach myself lessons that I used to have to go to Asia to learn.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #96 Posted: Wed Dec 11, 2019 1:58 pm 
Honinbo

Posts: 8903
Liked others: 1509
Was liked: 1526
KGS: Kirby
Tygem: 커비라고해
John Fairbairn wrote:
I think this discussion has reached a point where we start going round in circles. At least I feel it's time for me to leave it.

FWIW what I go away with is a feeling that the debate can be summed up as follows:

1. Side A: We can learn from AI but it will take a long time. Side B: Agreed.

2. Side B: We can learn from humans and can learn NOW. Sheldonian Side A: Disagree! How can a bunch of frightened dinosaurs even begin to think not using science can possibly work? Their approach has only got them to 9-dan level. Almost as bad as being an engineer from MIT! And that's my spot.

My vote goes to Penny.


An argument similar to Side B could be made as to why we should try to learn from our amateur club members, and not from professionals. After all, for many of us, folks at our local go club are more accessible than professionals. And if you're not the strongest person in the club, you could learn from them.

Same thing with the earlier post here:
Quote:
I find the argument about being shown a new move by a bot as a reason to value bots spurious.

Yes, it's interesting, if you like that kind of thing. Yes, it probably is a better move.

But you get that kind of insight over and over again in playing over human pro games. With the humans you also, in many cases, get a useful human commentary as to why it is a good move.

In fact, over the years the one thing that has improved my game much more than anything else is playing over a pro game and seeing a move to which my reaction is: "Oh, I didn't know you could do that!"


You could say, "I find the argument about being shown a new move by a pro as a reason to value professional opinion spurious. Yes, it's interesting, if you like that kind of thing. Yes, it probably is a better move. But you get that kind of insight over and over again in playing over amateur dan player games. With our amateur dan playing friends you also, in many cases, get a more personal commentary as to why it is a good move."

---

It is true that it's a little bit harder to talk to computers than it is to talk with humans. But for folks that aren't living in Asia or going to pro events, it's also a little bit harder to get insights from pros, when compared to getting insight from your amateur friends. Yeah, a professional go player might write a book or two. But what about all of those questions you have that the pro hasn't covered?

Generally speaking, bots can play better than pros, and pros can play better than our amateur go-playing friends. If you're interested in learning, the ideal would be to have perfect communication with the top bot so that it could explain everything to you. We're not really there yet. It's a lot easier to talk to folks who are more accessible - even to ask questions here on L19 to get amateur opinions. If you ask questions here on L19, there are sure to be mistakes in the responses. It doesn't mean that it's worthless to ask - you might learn something nifty, anyway. If you want more accuracy, you should go higher up the skill ladder - ask a pro if needed. If you need more accuracy than that, a bot might be able to point out pro mistakes.

---

I personally think it's all useful - asking questions here on L19, talking with pros, playing pros, analyzing stuff with bots. We don't have to have one or the other. But you'll be getting different benefits from different mediums. Bots are very high on the accuracy scale. A little bit lower on the explanation scale. L19ers may be able to talk to you all day. But we might not be that accurate in everything we write :-D

_________________
be immersed

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #97 Posted: Wed Dec 11, 2019 2:32 pm 
Lives in gote
User avatar

Posts: 302
Liked others: 52
Was liked: 69
Rank: EGF 5k Foxy 2k
TelegraphGo wrote:
John Fairbairn wrote:
I think this discussion has reached a point where we start going round in circles.


Agreed - since it feels like people have founds the positions they feel comfortable with, so should we now actually make a decision on what to do about old SL pages? Perhaps we should start a poll, or series of polls, with a bunch of options ranging from the extremes of rewriting with primarily AI analysis to leaving them 100% alone?

I find it fascinating and highly peculiar that anyone would choose not to consult Go Seigen's opinion. Gossip can be found everywhere and anywhere, Go Seigen's opinion cannot. It feels like opting not to take a free shot on goal - sure, the goalie might block it, but what if they don't? It's likely none of us would understand what he had to say, but what if some small part of us did, just a little bit more than before? I'd pay money per move to show him specific positions and hear him say no more than "L8". Luckily, I don't have to pay, because Leela is free! A few robotic tendencies are easy enough to ignore, when I can actually teach myself lessons that I used to have to go to Asia to learn.


Isn't it interesting that Go Seigen was fond of two space high enclosures, attachments, and paid much attention to who had sente. Remind you of anything? He was standing above everyone else quite clearly.

My opinion about the Sensei's Library pages: run them through strong AI. If the variation is clearly wrong, then sorry, it is wrong. And, if you meet an opponent who knows the refutation, then the opponent will have gained. That much to me is very clear. Therefore, those articles should have a clear warning at the top of the page pending a rewrite or a change of circumstances on the board that AI agrees with. As for advice (words), well, it is a community wiki. Advice can be subjective. But as a minimum, the variations must stand up to scrutiny!

I was discussing with a friend at the club the following which I will leave with you to ponder: we have had sweeping changes in Go theory, from old times where corners were left empty and 4-4 was not a move, to new fuseki, to the 70's where the taisha, magic sword and avalanche were so popular, and now we have AI. The question is: will there be any more sweeping changes? Computers have already told us what to play. Maybe we have reached the destination. This is, broadly, how to play Go.

_________________
12k: 2015.08.11; 11k: 2015.09.13; 10k: 2015.09.27; 9k: 2015.10.10; 8k: 2015.11.08; 7k: 2016.07.10 6k: 2016.07.24 5k: 2018.05.14 4k: 2018.09.03 3k: who knows?

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #98 Posted: Wed Dec 11, 2019 2:49 pm 
Lives in gote

Posts: 347
Location: Adelaide, South Australia
Liked others: 137
Was liked: 173
Rank: Australian 2 dan
GD Posts: 200
John Fairbairn wrote:
In real life, those old wives were your mothers and grandmothers, and they did a pretty good job of getting you where you are today.

Don't forget the curmudgeonly grandpa characters, reminding young people today that they understand nothing. When it becomes time for me to play that part, I'll be glad I had such good role models :-)

Honestly John, it's great to have you here, stopping us kids from getting too complacent.
John Fairbairn wrote:
It's long lane that has no loaf on the bread.

Thanks, this has made my day!

Top
 Profile  
 
Online
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #99 Posted: Wed Dec 11, 2019 5:22 pm 
Honinbo

Posts: 9417
Liked others: 2945
Was liked: 3150
John Fairbairn wrote:
I think this discussion has reached a point where we start going round in circles. At least I feel it's time for me to leave it.

FWIW what I go away with is a feeling that the debate can be summed up as follows:

1. Side A: We can learn from AI but it will take a long time. Side B: Agreed.

2. Side B: We can learn from humans and can learn NOW. Sheldonian Side A: Disagree! How can a bunch of frightened dinosaurs even begin to think not using science can possibly work? Their approach has only got them to 9-dan level. Almost as bad as being an engineer from MIT! And that's my spot.

My vote goes to Penny.


My view is this. We can learn from AI now, and we already have done so. We can learn from humans now, and many of us have already done so. Some of what we learned from humans we now have to unlearn, and the sooner the botter. (I intended to write better, but I like the pun. ;))

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins


This post by Bill Spight was liked by: Applebaps
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #100 Posted: Sun Dec 15, 2019 5:32 am 
Gosei
User avatar

Posts: 1516
Location: Ghent, Belgium
Liked others: 218
Was liked: 681
Rank: KGS 1d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
OK, here's another one, which is too interesting not to discuss:

We have the SL page "Learning joseki loses two stones in strength" : https://senseis.xmp.net/?LearningJoseki ... esStrength

It warns against blindly playing joseki and always carefully thinking about the meaning of your moves. This is important advice which is here to stay.
The page features an example by the famous professional Fujisawa Hideyuki:


Click Here To Show Diagram Code
[go]$$W Joseki but kikasare
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . 7 . . . . 5 . . . |
$$ | . . X O O O . . . . . . . 1 . . 6 . . |
$$ | . X . X . . X . . , 2 . . . . X . . . |
$$ | . . . X O O O . . . . . . 3 . . . . . |
$$ | . . . . X X . . . . a . . . . 4 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


Fujisawa is quoted to say that the choice of :w5: and especially :w7: is not good in this position. White should attack the black stone at the top with A. The result is "kikasare", i.e. White is forced, by her own choices, into a low position and :b2: is turned into a kikashi/forcing move/reducing move.


Click Here To Show Diagram Code
[go]$$W Active play
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . 5 . . . |
$$ | . . X O O O . . . . . . . 1 . . 6 . . |
$$ | . X . X . . X . . , 2 . . . . X . . . |
$$ | . . . X O O O . . . . . . 3 . . . . . |
$$ | . . . . X X 7 . . . 9 . . . . 4 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O 8 . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


The page recommends to lean with :w7: and then cap with :w9:

The advice makes total sense. In an influential position, you surround, rather than undercut. Go is the surrounding game.

The problem is that after many playouts LZ doesn't support this view.


Click Here To Show Diagram Code
[go]$$W Taking sente
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . 1 . . . O . . . |
$$ | . . X O O O . . . . . . . O . . X . . |
$$ | . X . X . . X . . , X . . . . X . . . |
$$ | . . . X O O O . . . . . . O . . . . . |
$$ | . . . . X X . . . . . . . . . X . . . |
$$ | . . . . . . . . . 2 . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 3 . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


First of all, what LZ really wants to do here is be the first to play in the open asymmetric corner. So she settles the top with :w1:, which apparently makes :b2: urgent and so she can turn to the lower right with :w3:

:w1: is not usually on our radar. I paused at 6,3k and 54,6% for White


Click Here To Show Diagram Code
[go]$$W Playing actively
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O . . . |
$$ | . . X O O O . . . . . . . O . . X . . |
$$ | . X . X . . X . . , X . 2 . . X . . . |
$$ | . . . X O O O . . . . . . O . . . . . |
$$ | . . . . X X 1 . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


Now to the leaning maneuver. LZ doesn't think this is locally sente and has Black counterattack with :b2:. I won't post the lengthy preferred sequence here but it leads to White sacrificing the top stones to build a center.

After 7,3k playouts, this counterattack gives Black 54%

So, with respect to LZ's preferred slide, this diagram 8,6% worse (by LZ's probabilities) and there's an effective switch in who's the favorite.


Click Here To Show Diagram Code
[go]$$W Playing actively
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O . . . |
$$ | . . X O O O . . . . . . . O . . X . . |
$$ | . X . X . . X . b , X . a . . X . . . |
$$ | . . . X O O O . . . . . . O . . . . . |
$$ | . . . . X X 1 . . . 3 . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O 2 . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


What if we grant White :w1: to be sente and feed Fujisawa's advice to the bot? Then Black will sabaki starting with A then B. As a result, White will have influence to tackle the right side. The evaluation is Black 49,8% after 7k playouts.

The result is better but still with respect to LZ's preferred slide, Fujisawa's lean & cap is 4,2% worse (by LZ's probabilities) and instead of WHite being the favorite, the chances are now even.


Click Here To Show Diagram Code
[go]$$W Kikasare again
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . 1 . . . . O . . . |
$$ | . . X O O O . . . . . . . O . . X . . |
$$ | . X . X . . X . . , X . . . . X . . . |
$$ | . . . X O O O . . . . . . O . . . . . |
$$ | . . . . X X 3 . . . . . . . . X . . . |
$$ | . . . . . . . . . 2 . . . . . . . . . |
$$ | . . . O 4 . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 5 . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


Now to the "catenaccio joseki" as SL has baptized it. LZ develops two different ideas for Black next. The first, as above, is very similar to her original idea with the smaller slide, taking sente to play :w5:. After 6k playouts, White has 53,6%. Almost equivalent (with less playouts) is to pay :w3: at :w5: directly. The big slide, with this follow up, is marginally worse (1%) than the top choice of the small slide.

The second is more difficult to interpret.


Click Here To Show Diagram Code
[go]$$W Kikasare again (2)
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . 1 . . . . O . . . |
$$ | . . X O O O . . . . . . . O . . X . . |
$$ | . X . X . . X . . . X . . . . X . . . |
$$ | . . . X O O O . . 5 . . . O . . . . . |
$$ | . . . . X X 3 . 2 . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O 4 . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 6 O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


This more forceful variation shifts Black's attention to the lower left with :b6: for reasons I don't understand. Given the variations, there are relationships between the elephant's eye, sacrificing it to link up at the top, and playing on the left side.


%%%%%

What to take away from this analysis?


To rehash the metrics:
- LZ's top choice, the small slide, gives White 54,6%
- the "blindly played catenaccio joseki" which Fujisawa scorns, gives White 53,6%
- Fujisawa's recommendation to play actively, leaning and capping, gives White 50,2%
- and if Black counterattacks, ignoring the lean, the result even drops to 46%

Of course we can say, what works for Fujisawa works for me, or we shouldn't care about the crazy things LZ says, we can still learn from conventional wisdom, carefully developed knowledge by professionals, crystallized into clear guidelines. And if we play by yesterday's pros standards, we are still at least 6d today. And as there is more such material, it's probably easier or better for us to learn that consistent path, rather than trying to figure out what the bots tell us.

But the things LZ tells us are not all over the place. Yes, there's a diagram here which suggests implications that I don't understand. But otherwise they are consistent with each other and consistent with other bots' stories. They tell a different story than the one we have developed by interpreting conventional wisdom for better or worse. Not "surround", "attack for profit (a potential profit being influence") ... but "settle fast", "play elsewhere asap", ...

The capping play in particular is under scrutiny. We know it blocks the path to the centre. It may force our opponent to live small while we take influence. The bots see that as a lost opportunity to take territory (where the opponent lives) and don't value influence as universally as (we understood) before.

Myself I may still lean and cap. But I find it interesting that the bots say: settle and play elsewhere.


This post by Knotwilg was liked by: xela
Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 113 posts ]  Go to page Previous  1, 2, 3, 4, 5, 6  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group