It is currently Thu Oct 31, 2024 4:29 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 141 posts ]  Go to page Previous  1 ... 3, 4, 5, 6, 7, 8  Next
Author Message
Offline
 Post subject: Re: How to Become a Dan
Post #101 Posted: Fri Aug 31, 2018 9:24 am 
Oza
User avatar

Posts: 2420
Location: Ghent, Belgium
Liked others: 359
Was liked: 1020
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Lecture 2, in which LZ finds a better local move than Kim 9p's joseki and also casts some doubt on the firm difference Kim sees with the "non-joseki"


Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #102 Posted: Fri Aug 31, 2018 9:39 am 
Oza
User avatar

Posts: 2420
Location: Ghent, Belgium
Liked others: 359
Was liked: 1020
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Analyzing lecture 3 with LZ gives the most astonishing result until now: here's a non-joseki all of us know as a beginner mistake which can be duly punished. LZ agrees that the joseki move is better but not for the shape consequences we have been taught but because ... it keeps sente and not just that.

I think I've discovered a new proverb: "use sente to play 3-3 in the corner opposite to the current action"


Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #103 Posted: Fri Aug 31, 2018 5:50 pm 
Lives in sente

Posts: 758
Liked others: 114
Was liked: 916
Rank: maybe 2d
Interesting that you hadn't seen the following move before, this way of refuting this sequence I've definitely seen pros lecture about and teach, actually slightly more frequently in my personal encounters I've seen this way taught than the other way. I had thought it was pretty standard knowledge (along with, for dan players, considerations about if white has a top side stone that could let him try E18 cut and then crawl at F18).

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


I do think different pros have in the past favored teaching different sequences in response to the initial hane. Having seen both ways taught, I'm glad to see LZ at least in this one case suggest the one I personally favor a bit more. :)

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #104 Posted: Fri Aug 31, 2018 9:28 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Click Here To Show Diagram Code
[go]$$Bc
$$ ------------
$$ | . . . . . . . . .
$$ | . . . . . . . . .
$$ | . 1 X X W . . . .
$$ | . . O O 2 . . . .
$$ | . . . . . . . . .
$$ | . . X . . . . . .
$$ | . . . . . . . . .
$$ | . . O . . . . . .
$$ | . . . . . . . . .[/go]


When I was 3 dan in game reviews I started telling my opponents who had played the hane, :wc:, to connect at :w2:. I even played that hane-and-connect sometimes myself. ;) I had realized that in general the connection was good, but I still thought that the descent, :b1:, was correct.

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


Until AlphaGo, I still thought that :w1: was correct, even though :bc: still had significant aji. To counter that I usually continued with :w3:, expecting :b4: in response, and happy when Black did not play it.

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


IMX, most beginners wanted to connect underneath with :w1:. I told them that was wrong. :oops: OC, :bc: still has aji because of the threat to cut, but it's not too hard to convince yourself that the beginners were right.

Later, when I learned about this clamp, the argument seemed rather subtle to me.

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


Tewari comparison with joseki indicated that the exchange of :wc: and :b3: was aji keshi. We now have to recognize that :ws: is misplaced, as well. ;)

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

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #105 Posted: Sat Sep 01, 2018 12:02 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
RE lecture 3: Yes, I'm rather surprised Knotwilg didn't know the descent being the fake punishment for DDKs nor seen the clamp! I've been saying that on these boards for years when this hane not extend question is asked, as it repeatedly is. See viewtopic.php?p=161728#p161728 from 2014 and following dozen posts with discussion and links to other threads from 2014 and 2012.

P.S what LZ and how many playouts are you using Knotwilg? The way you write about variations ending suggests you might be just looking at principle variation of suggested move instead of playing them out. For example for me LZ #157 in lecture 2 wants to crawl not cut after the shoulder hit joseki from 1k to 50k playouts by 55% vs 52%. And the classic punishment for white's cut in the case black has the ladder is to atari on the outside and push, a move LZ barely considers. But that could be revealing a weakness of LZ as it's bad at seeing ladders 10 moves in the future. So if I take the role of black "punishing the cut" and let LZ play as white let's see what happens. It sees the ladder within a few thousand playouts and we end up with black capturing the 2 stones (with some bad aji) and white getting a wall outside with black at 47%, lower than the push down is predicted to be at that time (but actually playing it out black's win% deteriorates, interesting it doesn't play the supposed tesuji cut):
Attachment:
llknot-l2a.PNG
llknot-l2a.PNG [ 808.79 KiB | Viewed 35866 times ]


P.P.S we should remember the lecture series is called "How to Become a Dan" so is aimed at kyus. So recommending the descent instead of hane, which is IMO (and not-ancient pro opinion) a bad move unless your opponent is a generous noob and blocks is fair to criticise as bad teaching for being outright wrong. But something like tenuki or the counter pincer instead of shoulder press in lecture 2 is fine teaching as it's a simpler move that makes your stones work together (and only -0.2% in LZ 157). Pros do sometimes play a counter pincer (with a little more global reason to do so). LZ doesn't know how to temper its advice from "How to Become a 10p".

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #106 Posted: Sat Sep 01, 2018 1:08 am 
Oza

Posts: 3698
Liked others: 20
Was liked: 4660
Click Here To Show Diagram Code
[go]$$Wc
$$ ------------
$$ | . . . . . . . . .
$$ | . . . . . . . . .
$$ | . X X X . 2 4 . .
$$ | . X O O O . . . .
$$ | . O 1 . . . 3 . .
$$ | . . B . . . . . .
$$ | . . . . . . . . .
$$ | . . O . . . . . .
$$ | . . . . . . . . .[/go]


I'm puzzled why both knotwilg and Bill seem to have regarded this as joseki. Kitani, in his joseki dictionary, says it is bad for White as he is overconcentrated.

In fact Kitani says E16 was the mistake. He recommended B17, which doesn't even get a mention in my LZ setup. LZ offers only E16 and the B14 link-up, which Kitani in turn doesn't mention. (However, all Kitani's judgements should probably be seen as being based on no-komi go).

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #107 Posted: Sat Sep 01, 2018 8:07 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
John Fairbairn wrote:
Click Here To Show Diagram Code
[go]$$Wc
$$ ------------
$$ | . . . . . . . . .
$$ | . . . . . . . . .
$$ | . X X X . 2 4 . .
$$ | . X O O O . . . .
$$ | . O 1 . . . 3 . .
$$ | . . B . . . . . .
$$ | . . . . . . . . .
$$ | . . O . . . . . .
$$ | . . . . . . . . .[/go]


I'm puzzled why both knotwilg and Bill seem to have regarded this as joseki. Kitani, in his joseki dictionary, says it is bad for White as he is overconcentrated.

In fact Kitani says E16 was the mistake. He recommended B17, which doesn't even get a mention in my LZ setup. LZ offers only E16 and the B14 link-up, which Kitani in turn doesn't mention. (However, all Kitani's judgements should probably be seen as being based on no-komi go).


If I had to guess, I'd say it was from an article in Igo Curabbu many moons ago. Since my way of studying joseki is mainly to look something up if I feel I have messed up, I don't think I ever had occasion to look this sequence up. :lol:

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

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #108 Posted: Sun Sep 02, 2018 6:15 pm 
Oza
User avatar

Posts: 2408
Location: Tokyo, Japan
Liked others: 2346
Was liked: 1332
Rank: Jp 6 dan
KGS: ez4u
John Fairbairn wrote:
Click Here To Show Diagram Code
[go]$$Wc
$$ ------------
$$ | . . . . . . . . .
$$ | . . . . . . . . .
$$ | . X X X . 2 4 . .
$$ | . X O O O . . . .
$$ | . O 1 . . . 3 . .
$$ | . . B . . . . . .
$$ | . . . . . . . . .
$$ | . . O . . . . . .
$$ | . . . . . . . . .[/go]


I'm puzzled why both knotwilg and Bill seem to have regarded this as joseki. Kitani, in his joseki dictionary, says it is bad for White as he is overconcentrated.

In fact Kitani says E16 was the mistake. He recommended B17, which doesn't even get a mention in my LZ setup. LZ offers only E16 and the B14 link-up, which Kitani in turn doesn't mention. (However, all Kitani's judgements should probably be seen as being based on no-komi go).


I was interested because some of Kitani's opinions were quite iconoclastic. AFAIK (via a quick review in GoGoD) he never played this joseki as White (either with E16 or B17). In fact he does not seem to have liked the low 1-space pincer in general. The only game in which he played it was against Kubomatsu in the 1935 Oteai. Kubomatsu answered with a double approach at F17 rather than jumping in to 3-3.

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

Top
 Profile  
 
 Post subject: Re: How to Become a Dan
Post #109 Posted: Tue Feb 12, 2019 2:02 am 
lovelove wrote:
This will be about common mistakes usually seen in kyu games, and how to punish them.
Given the extraordinary popularity of this thread, only a very foolish person would dare to risk her life by daring to question the very basis of the entire lecture series.

So here we go ... :)

How to become strong is, i humbly submit, the exact opposite of learning how to punish weak players' mistakes, for all that will do is turn you into the Neighbourhood Bully that Bob Dylan sang so eloquently about.
To hear the song the way it's meant to be heard, download it into your local video editor and slow it down!

So - Instead of beating up smaller kids, pit your wits against those stronger than you, and aim to learn from the way they handle your own blunders.

Of course, it won't be apparent to you which of your moves is a blunder, but Lizzie or Sabaki can tell you when you are safe back home with a copy of your sgf (or a video of your game) in your pocket.

Top
  
 
Offline
 Post subject: Re: How to Become a Dan
Post #110 Posted: Mon Jan 31, 2022 3:41 pm 
Oza
User avatar

Posts: 2420
Location: Ghent, Belgium
Liked others: 359
Was liked: 1020
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
I have evaluated a few more of these lectures with KataGo and I can say they are essentially rubbish.

The recommended move is often just as doubtful as the "mistake", often the best move is not taken into account, or the best move is only marginally better. On occasions there's a move in a sequence which KataGo drops 8-10 points while it seems to be incidental to the explanation given.

I don't blame the original author or the professional giving the lectures for their genuine efforts, but these lectures "How to become a Dan" seem to have no ground in actual positional evaluation.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #111 Posted: Tue Feb 01, 2022 9:59 am 
Oza

Posts: 2495
Location: DC
Liked others: 157
Was liked: 443
Universal go server handle: skydyr
Online playing schedule: When my wife is out.
Knotwilg wrote:
I have evaluated a few more of these lectures with KataGo and I can say they are essentially rubbish.

The recommended move is often just as doubtful as the "mistake", often the best move is not taken into account, or the best move is only marginally better. On occasions there's a move in a sequence which KataGo drops 8-10 points while it seems to be incidental to the explanation given.

I don't blame the original author or the professional giving the lectures for their genuine efforts, but these lectures "How to become a Dan" seem to have no ground in actual positional evaluation.


If they were in line with professional evaluation at the time, I think it's a bit disingenuous to claim that now that we have better analysis, their older analysis was useless. I think it's worth treating them not so much as 'this is what to do in this exact situation' as 'this is how strong players from this time period conceive of the game and what you should consider aiming at'. As you can see from the earlier commentary, none of the stronger players on here seemed to think there were any major problems with any of the lectures at the time.

Of course, the key is recognizing that these are 10 years old and pre-AI.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #112 Posted: Tue Feb 01, 2022 11:19 am 
Lives in sente

Posts: 967
Liked others: 24
Was liked: 173
Rank: panda 5 dan
IGS: kvasir
I think the problem with these lectures is not that they are "wrong". I think the problem is two fold: the subject matter is some too common patterns that maybe can be avoided (boring?), and secondly it is a superficial analysis.

I have seen some of these topics covered in 30-60 minute lectures by professionals and you wouldn't really complain that the premise, that some move is a mistake, was wrong because the lecture is in much more depth. Lecture 2 is one position that I have seen lectures on and it was very instructive to consider the life and death of the white group, the difference between a nobi and a 1 space extension, the possibility of black taking the corner, some example professional games (from the 90s, pros haven't played like this for a long time) and so on. Nothing like that is covered in this lecture and all of these lectures amount to only showing a very small number of variations as if your opponent is going to go along with this.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #113 Posted: Tue Feb 01, 2022 5:20 pm 
Oza
User avatar

Posts: 2420
Location: Ghent, Belgium
Liked others: 359
Was liked: 1020
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
I don't have a problem with the fact that professional insights pre-Ai were skewed towards influence while AI shows an even bigger preference for corner territory. AI has an idiosyncratic way of dealing with moyos which is or was beyond professional comprehension. 5-4 points, pincers ... have fallen out of grace and the immediate 3-3 invasion has come into fashion. The common pattern after a close pincer of the 6-3 approach has changed: the second line crawl has dethroned the 3rd line connection. All of that is "forgivable".

This series was aimed at aspiring dan amateurs, showing clear "mistakes" and clear "best moves", distilled from pro play which was above discussion, only the variations are discredited by AI to an extent that the analysis and narrative can be dismissed as almost random comments and unsubstantiated claims. I find that a little disappointing.

I had downloaded this series as a reliable source of information, giving a number of standard sequences in or right after the opening. I have now removed it from my laptop. It's essentially worthless.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #114 Posted: Tue Feb 01, 2022 7:51 pm 
Honinbo

Posts: 9550
Liked others: 1602
Was liked: 1712
KGS: Kirby
Tygem: 커비라고해
It's a bit strong to call the information worthless.

KataGo is a lot stronger than pros, but it still doesn't know how to play go optimally. For the sake of argument, you can imagine a super-AI 20 years in the future that is significantly stronger than today's KataGo. That super-AI may make today's KataGo moves appear to be "rubbish".

But we know that today's KataGo moves are NOT rubbish; we learn from them, and can get new ideas. The same is true of this series, even if a stronger AI refutes the recommended sequences.

In my case, if I hear a recommendation from a 5-dan amateur, I'll think about it and probably respect the thought process, even if KataGo disagrees with it. Not because the 5-dan amateur's thought process is stronger than KataGo's. Rather, because that 5-dan amateur's thought process is stronger than my own.

_________________
be immersed

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #115 Posted: Wed Feb 02, 2022 3:24 am 
Oza

Posts: 3698
Liked others: 20
Was liked: 4660
Quote:
KataGo is a lot stronger than pros, but it still doesn't know how to play go optimally.


I agree with the gist of your wider comments, but I think we can go even a bit further, in particular with the above statement.

I would suggest that this could be rephrased as "AI is more successful than pros." Not stronger, because when we say someone is strong we tend to be implying they know a lot about the game. Cho Chikun is VERY strong but is not specially successful at the moment.

I would go further then and say that bots are more successful mainly because they don't get tired or lose concentration or feel pressure or get bad habits. Were a human able to eliminate all these frailties, I think we would find that they are very, very close to the bots. I say this partly because I've been watching various grandmasters commenting on the Tata Chess Tournament. I was struck by the fact that they now rarely mention "the engines" and, when they do, they are often dismissive of them (e.g. even the engines can't play this endgame properly without a huge amount of time). I've already mentioned here that we are seeing something similar in pro go. There was a Big 7 title match recently (Kisei Game 2) which had a long commentary. AI was mentioned just once, and that was to check correct play in a tricky capturing race (and the pro play was confirmed as correct). What I sense is that pros have long got over their initial AI shock and are now feeling more assertive and confident about their own views and knowledge. If that is justified, and I think it is, we are justified in still cleaving to what they tell us.

I think there is a simple but telling experiment that should convince doubters. I'm sure most people have even tried it already. Play a bot and get creamed. Then play a bot but after every stupid move by you, take your move back and try another. By "stupid" I mean a move where, as soon as the bot plays you realise instantly that you fell into a hole because e.g. you carelessly filled in a liberty. You were careless, not ignorant. You know the tesuji but you didn't look. You played like a typical human. And when you take back your moves and play more carefully, you will suddenly discover that you are maybe four or five stones stronger.

Go further and imagine that you had an app that flashes up the value of every boundary play on the board (2 points gote here, 5 points reverse sente there, and so on). You KNOW how to count these positions, but in practice you can't be bothered with all that fiddly stuff, so you play by guessing. But of the values were displayed, you could probably play a very decent endgame. Accepting Rin Kaiho's argument that amateurs lose abut 30-40 points in the endgame, you would find that you've suddenly become 2 to 3 stones stronger - with no extra knowledge.

This is not so strange. After all, top tennis players hire teams of people to help them prepare. These people are not as good at hitting the ball as the player, so can't really teach the player anything about the mechanics of tennis. They are there to provide motivation, detect bad habits, plan tactics, keep the player fit, relaxed etc etc.

Go pros don't have this support, except in China for international events. Amateurs don't have it and have to make do and mend. I believe go (and chess) amateurs who study with bots, who devise ways of counting thickness, who memorise josekis or even whole games, or who cuddle their cats or do Pilates while they sit at the go board are all, in a sense, chasing fool's gold. At least when they tell themselves things like "I'm now stronger because bot X shows me my main mistakes." That is not to say that I think such effort is wasted. Far from it. It's just that I think the cause-and-effect is not what they think it is. I believe the true cause of any improvement is simply more time spent at the go board. Whatever floats your boat to make you spend that time is good for you, as an individual. Other people probably need different motivations. Some people told they need more exercise may simply buy a treadmill. Others may buy the latest Nike trainers, which makes them go to the gym so they can be seen with their new bling. Once there they go on the gym's treadmill. Some may get even swankier and hire personal trainer at the gym. They all get fitter. They may feel it's their new shoes or the hunky personal trainer that made the difference, but in each case it's the time on the treadmill that deserves the credit.

Do bots help go players with motivation? My impression is that they may help pros, because they make very few mistakes and these can be pinpointed by the bot. The task of rectifying the mistake is then doable. But amateurs make lots of mistakes and so are overwhelmed by the bots. I remain to be convinced that having a bot point out you made a 10-point mistake is of any real value, when each of the previous ten moves you made all made a 1.5 point loss. The real task you face, both as a pro and an amateur, is to work out WHY you make mistakes, not WHAT the mistakes are, because the odds are that you already have knowledge of the WHAT (in the game I referenced above, Ichiriki said straight after the game something like "I know I made a lot of mistakes in the endgame" - he didn't need a bot to tell him that). Bots are incapable of telling you WHY you made a mistake. Unfortunately it seems that humans are likewise not very good at telling you as an individual WHY you make mistakes. It may be for entirely non-go reasons. E.g. you are unfit and so can't concentrate for long periods.

One reason I deplore the lack of discussion and anonymity on L19 is that go is by and large a solitary activity at the best of times. In the modern age, with people playing in front of computers instead of people, and hiding behind handles and avatars, they are becoming more and more like those people what just watch porn in their hotel room when they travel. It really is a lot more fun to get out and mix with the locals. And finding out what makes the local people tick is actually a better way to true knowledge than listening to a guide who tells you this cathedral was built in 1573. That's because, if you understand how other people tick, you are more likely to realise what will make you tick. Ergo, people are better for you than bots (or tour guides).


This post by John Fairbairn was liked by: gowan
Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #116 Posted: Wed Feb 02, 2022 5:24 am 
Oza
User avatar

Posts: 2420
Location: Ghent, Belgium
Liked others: 359
Was liked: 1020
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Kirby wrote:
It's a bit strong to call the information worthless.


It's indeed strong. It does reflect my honest feeling though.

Quote:
KataGo is a lot stronger than pros, but it still doesn't know how to play go optimally. For the sake of argument, you can imagine a super-AI 20 years in the future that is significantly stronger than today's KataGo. That super-AI may make today's KataGo moves appear to be "rubbish".

I don't expect super AI to qualify today's choices by KataGo a 10 point mistake. Just like I don't expect an AI ever to prove that the L-group is not dead. I'm not saying any human knowledge is rubbish, I'm saying this particular one is, and I wonder how it could get to that.

Quote:
In my case, if I hear a recommendation from a 5-dan amateur, I'll think about it and probably respect the thought process, even if KataGo disagrees with it. Not because the 5-dan amateur's thought process is stronger than KataGo's. Rather, because that 5-dan amateur's thought process is stronger than my own.


A priori, yes. When KG points out 10-point mistakes in a diagram, incidental to the point the author is making, which is rather invalid as well, I'm getting suspicious.

This series is not "how professionals evaluate difficult positions today" but "how to become dan". For that reduced purpose, the lectures go way out of their league, representing variations as "mistakes" and recommending best plays while it's actually the other way around. It's misleading, not intentionally so, but deep down the authors must have felt that "this outcome disrupts the san ren sei" was a vague argument to dismiss the choice KG actually favors.

I'll make another tricky comparison. At the start of the pandemic I trusted the clinical experts 100%. Seeing how their recommendations and findings were invalidated by reality time and time again, the authority argument made room for evidence, or lack thereof. As a scientist, I have to trust experts to start with but I should always be ready to challenge their expertise when it is defied by the scientific method.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #117 Posted: Wed Feb 02, 2022 6:56 am 
Oza
User avatar

Posts: 2420
Location: Ghent, Belgium
Liked others: 359
Was liked: 1020
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
John Fairbairn wrote:
What I sense is that pros have long got over their initial AI shock and are now feeling more assertive and confident about their own views and knowledge. If that is justified, and I think it is, we are justified in still cleaving to what they tell us.


By all means! But I doubt they will tell us the same story as the lectures in this series. I was not doubting human pro capacity to transfer purified expert knowledge to the masses. I was doubting the scope of these lectures and signaling what seems to have been an overestimation of said insight.

Quote:
And when you take back your moves and play more carefully, you will suddenly discover that you are maybe four or five stones stronger. Accepting Rin Kaiho's argument that amateurs lose abut 30-40 points in the endgame, you would find that you've suddenly become 2 to 3 stones stronger - with no extra knowledge.


I fully agree. As you have pointed out, I erroneously labeled these things as "gamesmanship", but indeed, having the lucidity and discipline to select good candidates, read, evaluate the position and remain aware of changing conditions, sleep well, don't drink or fiddle with your smartphone while playing ... will have a bigger sudden impact on your playing level than any game review with AI or new book on the shelf. Applying yourself during the game takes it to its potential level. Studying raises the potential level.

Quote:
Do bots help go players with motivation?


They help with mine :). I can't tell for others.

Quote:
One reason I deplore the lack of discussion and anonymity on L19 is that go is by and large a solitary activity at the best of times. In the modern age, with people playing in front of computers instead of people, and hiding behind handles and avatars, they are becoming more and more like those people what just watch porn in their hotel room when they travel. It really is a lot more fun to get out and mix with the locals. And finding out what makes the local people tick is actually a better way to true knowledge than listening to a guide who tells you this cathedral was built in 1573. That's because, if you understand how other people tick, you are more likely to realise what will make you tick. Ergo, people are better for you than bots (or tour guides).


It's been a while since I last saw a hotel room, let alone watched porn in that setting, nor did I recently go out to mix with any other locals than those in my locality. I can relate to your argument when it comes to language, culture, geography, history ... and overall getting out is much healthier than sitting in front of the laptop. With Go I have a mixed experience. I never really enjoyed social Go events for the same reason I don't like table tennis tournaments spanning a weekend, or a guitar internship of a similar time span. I won't dwell on it but in general I prefer socializing with friends outside a specific domain, and I'm happy to keep the Go related exchanges in physical solitude.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #118 Posted: Wed Feb 02, 2022 7:34 am 
Honinbo

Posts: 9550
Liked others: 1602
Was liked: 1712
KGS: Kirby
Tygem: 커비라고해
Knotwilg wrote:
...

I'll make another tricky comparison. At the start of the pandemic I trusted the clinical experts 100%. Seeing how their recommendations and findings were invalidated by reality time and time again, the authority argument made room for evidence, or lack thereof. As a scientist, I have to trust experts to start with but I should always be ready to challenge their expertise when it is defied by the scientific method.


I generally agree with your skepticism (fwiw, I am also skeptical of clinical experts). I think the part where I differ in opinion is that I remain skeptical of AI, and don't see its evaluations to be much better for learning than that of pre-AI pros. Both are fallible, yet both are stronger than me. Because of this latter part, I think I can learn from both.

If I am actually as strong as a pro, at that point only can I appreciate the difference in opinion between AI and pros. Prior to that point, I'm just relying on what the AI tells me - and the general knowledge that AIs are stronger than humans.

--

For example, when you say, "When KG points out 10-point mistakes in a diagram...", you are trusting KG's evaluation of this. That is fine, but trusting this evaluation isn't based on scientific method so much as generally accepting the idea that "AI is strong". Scientific method would be more like considering the variation, then performing a test against that variation to see how it works out. One way to do this type of testing is to try it out in your own games. From that perspective, the variations described in the original lecture for this thread could very well be tested and validated by trying them out in your own games.

If you try out one of the variations, and it doesn't work out well for you... that's fine, move on. Maybe try an AI variation. But you should test that one, too. Try out the AI variation in your game. If that doesn't work out for you, too... Then move on, again.

Scientific method is testing these hypotheses to see if they work out for you. I think a great way to do that is by trying them out in your own games. For that reason, I think the original lecture and AI BOTH give useful information: a set of hypotheses, which I can test out in my own games. If I try them out and they're not working for me, then I can move on.

_________________
be immersed

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #119 Posted: Wed Feb 02, 2022 8:58 am 
Oza
User avatar

Posts: 2420
Location: Ghent, Belgium
Liked others: 359
Was liked: 1020
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Kirby wrote:
If I am actually as strong as a pro, at that point only can I appreciate the difference in opinion between AI and pros.
Prior to that point, I'm just relying on what the AI tells me - and the general knowledge that AIs are stronger than humans.
--
For example, when you say, "When KG points out 10-point mistakes in a diagram...", you are trusting KG's evaluation of this.
That is fine, but trusting this evaluation isn't based on scientific method so much as generally accepting the idea that "AI is strong".


There are a few proof points though, one of them being the 64-0 score by AI against top pros and the fact that pros are now using AI to check their own conceptions or learn new ones. Also, AI doesn't know (yet, or at least not my KataGo version) who's in front of them, so they are not trying to make any messaging easier to consume thereby making it wrong. KataGo shows me what I can see through Lizzie's interface and I have a few ways of interrogating it further. When persisting, I'm not wasting a pro's time or questioning his knowledge. The analysis, however correct, is more objective since it's voided of such emotions and fueled with exhaustive interaction.

Quote:
Scientific method would be more like considering the variation, then performing a test against that variation to see how it works out. One way to do this type of testing is to try it out in your own games.


That is exactly what I'm trying to do, although it happens at a slightly higher level: I'm trying to cure patterns in mistakes, because singular mistakes are very hard to reengineer. So I try to make less slow connections, surround in sente but not gote, be aware of superficial shape moves like the table shape, bullying attacking groups before submitting to defense ...

The standard moves upon which AI has improved, well, everybody plays them nowadays and their difference is not game deciding. The singular big mistakes in my games or in lectures like the above, can't be recreated.

Quote:
From that perspective, the variations described in the original lecture for this thread could very well be tested and validated by trying them out in your own games.

If you try out one of the variations, and it doesn't work out well for you... that's fine, move on. Maybe try an AI variation. But you should test that one, too. Try out the AI variation in your game. If that doesn't work out for you, too... Then move on, again.

Scientific method is testing these hypotheses to see if they work out for you. I think a great way to do that is by trying them out in your own games. For that reason, I think the original lecture and AI BOTH give useful information: a set of hypotheses, which I can test out in my own games. If I try them out and they're not working for me, then I can move on.


AI creates a huge sample of games to decide if a move is promising - my own sample will pale against that :)

But you have a point in that we shouldn't replace one blind belief with another. I'm definitely keeping the game as rich as possible by starting on 5-4 or pincering for example (or 3-3 next, or san ren sei ...). I believe there's even a competitive advantage nowadays of NOT following AI opening gospel.

Top
 Profile  
 
Offline
 Post subject: Re: How to Become a Dan
Post #120 Posted: Wed Feb 02, 2022 10:06 am 
Honinbo

Posts: 9550
Liked others: 1602
Was liked: 1712
KGS: Kirby
Tygem: 커비라고해
Knotwilg wrote:
There are a few proof points though, one of them being the 64-0 score by AI against top pros and the fact that pros are now using AI to check their own conceptions or learn new ones. Also, AI doesn't know (yet, or at least not my KataGo version) who's in front of them, so they are not trying to make any messaging easier to consume thereby making it wrong. KataGo shows me what I can see through Lizzie's interface and I have a few ways of interrogating it further. When persisting, I'm not wasting a pro's time or questioning his knowledge. The analysis, however correct, is more objective since it's voided of such emotions and fueled with exhaustive interaction.


I pretty much agree. However, the proof here proves overall game performance for pros against AI. There may be different things to learn for a local situation in weaker amateur games. Maybe both objectives lead to the same best move to play; maybe not. The reason I find it valuable to consider an idea and test in one's own game is because that type of testing eliminates one variable: you validate whether something works for you at your level, rather than trusting what worked for someone else (i.e. a superhuman level bot).

Ultimately, we all want to play as well as possible, so I can understand wanting to imitate the moves of the highest level player we have available (i.e. AI). However, on the road to achieve better play, there may be different things to learn and understand. I believe that these learnings can be acquired from players stronger than us, even if they are, e.g., amateurs.

Knotwilg wrote:

AI creates a huge sample of games to decide if a move is promising - my own sample will pale against that :)



That is certainly true. Your own samples have the benefit of being more personalized, however. You can feel the pain when something doesn't work; or you can feel the benefit when something does.

_________________
be immersed

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 141 posts ]  Go to page Previous  1 ... 3, 4, 5, 6, 7, 8  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