I appreciate the debate and the apparent interest in my progress as Aunt Sally for the Western go amateurs

But perhaps some the tangent on the value of language in the development of expertise deserves its own thread. It's not that I don't want to host the current discussion, so feel free to continue.

In the meantime, the promising results have not yet been confirmed. Last 5H game I lost by blundering in the late endgame while 13 points ahead but the good thing is that at some point I outplayed the 50 playout AI as confirmed by the 500 playout analysis AI.

https://neuralnetgoproblems.com/ seems to be down, so no training today there.

Strange, it's working fine for me right now.

I tried to create a new thread here haha.

Another win against 5H 50 playout AI Sensei. Not too many missed chances, still it reduced the 55 point advantage to below 15.
The danger of playing this for too long is that one becomes submissive. Almost any attempt to dictate the pace of the game (sente) runs into a counter. Most gote are either proper moves or not too bad.

On the other hand, it becomes very apparent how shape matters. White's stones are superefficient. Mine are often a little overconcentrated.


[/sgf]

Maybe move on to 4 stones....

My own experiences ("You" = me):

- At some point, you reach a handicap where you can't win by simply defending. It's worth exploring how KataGo would itself play as Black - which is surprisingly active.

- I think you learn a lot about "how not to lose points". A lot of 'active play' is an actually a form of gambling unless you can read it out at some basic level (obviously AI reading level is way above that). Overall, active play is a very good thing though.

- You develop a 'sixth sense' about vulnerabilities in groups/territories and where a future attack may come from further down the line.

- You develop a better endgame through repeated practice

- Completely beyond my understanding and abilities, there's unexpected levels of skill in emphasising the centre and applying pressure on groups, and through this being able conjure up moyos seemingly out of just a few moves. [Edit for clarity: obviously I'm already aware of this possibility, but this is with Black having 3 or 4 stones handicap already on the board ]

Thanks for the notes.

I think I'll move to 4 stones when the win/loss ratio gets closer to 1/2. Now it's about 1/4 and only 2 wins (8 losses).

AI sensei takes sente border plays (endgame) early, so as to stabilize in sente and take away some easy points.

And then there is the center indeed. One move that shows an improvement in my central play is 140. I would usually respond to the attachment (with a hane) but through previous games I saw the threat of my bigger group being isolated.

Overall I did rather well in the center this time and this was instrumental to victory.

Personally I find playing KataGo to be like solitaire. Without handicap it is too difficult but with some handicap it is much easier than playing strong human players.

In handicap games I find the following faults:
* it sometimes plays quickly into endgame positions without creating any complications. I think many players can win in those games if they are patient enough and have good enough endgame technique.
* it will sometimes play some large scale joseki that gives black solid outside while white takes territory. This is especially the case in higher handicap games and once you encounter these you have a good chance of repeating the whole sequence. With some practice it is possible to win this way.
* it repeats the same moves. That is it will play the same openings for many moves and it will have some joseki or shape that it will play too predictably. With some practice you can predict what it plays, it is not that you understand the position, you learn its personality.

When you change the weights it will change its playing style a lot but that doesn't seem to affect how it is sometimes too eager to get an early endgame. However, in Katrain there is an opponent called KataHandicap that is much better at complicating things in handicap games and more likely to try to trick black. I guess there is nothing wrong with a solitaire-like opponent, you practice some skill and then move on to a different handicap or different weights, but maybe KataHandicap is more similar to playing a strong player that is trying to win

Maybe large handicap games are always a test of patience. Still, I really feel like winning against KataGo with large handicap is more of a test of patience than skill.

Here is some evidence, my game just now with KataGo with 50 playouts, just my local game which hardly anything happened in except that my endgame was bad.

Interesting comments. What do you consider high handicap by the way (7- 9 stones?)

On your game:

I think it takes some level of skill to play patiently and solidly and make sure nothing untoward happens - whilst not losing too many points.

I couldn't see any obviously outright bad moves in the endgame. Were there any that you would point to?

If this isn't too much of a complete flight of fantasy (and hopefully not an unintentional insult), I can imagine a Pro looking at this game and saying "From the limited evidence of this single game, the solid style and the way Black managed the game, I'd estimate Black to be around European 4 Dan or so".

I agree playing too long H5 against AI can lead to bad habits. However, in the 20 years before, I hardly played any game higher than H2 as Black. I wanted to get advice that would force me into doing things I'd never done before, because that's going to get you out of a rut. Another such thing is arduous joseki study.

If there's any pattern in my losses, it's going into big fights where I have no idea if I will win. It's flipping a coin. Against AI, it's not flipping a coin, more like playing the state lottery. So I play more steadily, where all my moves are 0-4 pt errors. If I do that against an inferior player, I will more likely win. Over time I won't know how to fight anymore so I need to get back into more ambitious games. For now it's good healing practice.

I agree with the early endgame observation. They feel like thank you moves.

Thanks for the KataGo Handicap tip!

Edit: my practice program
- Thanks to the fantastic interface of AI Sensei, I have a training program of 83 problems now coming from my own games. When this gets to a coup
- I'm shifting from 1d problems to 3d on neuralnetgoproblems.com
- I'm doing daily tsumego via the mobile app Tsumego Pro

I am unhappy with moves 114 to 124 (I wanted to upload a diagram but there was some error, I guess you can refer to the sgf above). Most of the moves are OK but there are some bad ones.

I missed a tesuji for 110



After that I attached with 112 which feels useless but I did want to avoid reducing white's top group with a cuttable shape. Maybe giving up points like this was clever or maybe not.

Then 118 is not the right move. Black should hane and white should descend, if black descends it's same as if white played this area first. There was also time to play 120 once before hane.

Then 122124 can't be right. It is similar in size to 120 and 122 but those are sente and since 122124 is gote and not necessary I should have looked for another move.

Finally, I really wasn't playing the small endgames in the right order. I lost 6-7 points after 124 according to katago.


I don't know if that means anything to Asian pros but maybe European pros can think in this way When you pretext it with something about white being KataGo then that is one thing but without that information I think they could be confused about what this is

===Edit: Wrote 122 when I meant 124

Had a bit of a weird game against an unranked player on OGS. He resigned while ahead in the early middle game. He was stronger than I. His call to resign, I got another game to learn from with AI Sensei.

Now that dust's suggestion to play AI at 5H is starting to wear off, I'm accepting new challenges. I'm still onto John's suggestion to study his "nexuses" but take his advice not to do so while playing. Studying joseki seriously is another "getting out of my comfort zone" idea.

We can play a game together if you’d like?
I think from where you are at, I might need a two stone handi. Down to play even though.

_________________
My Baduk Journey

I have some suggestions - 1 and 2 are time consuming though.

1. Return to the physical tournament circuit. There’s something about playing long-ish games that matter that really focusses the mind and determination to improve.

2. Vocabulary alone and vague concepts won’t do it (imho). It might be an idea to create your own corpus (or nexuses/nexii?) of theory more explicitly and systematically (Robert is a good example of this in depth)


For non-time consuming ways I’d suggest:

3. push yourself out of the comfort zone to play many more higher rated players online.

4. Keep reviewing games with AI and focus on eliminating mistakes

The proper plural of nexus is nexus, believe it or not, but nexuses is nowadays deemed as acceptable. But that aside, I agree completely. Discovering things for yourself is the best way to learn anything. I've always found it strange that people use "you're re-inventing the wheel" as a negative criticism.

Reinventing the wheel is usually the only thing people not proficient in something can do, haha!

Thx dust.

1 is not an option today.
For 2: see https://senseis.xmp.net/?DieterVerhofst ... InMistakes
But I may want to grease those wheels
3 yes. That's the one
4 for sure

Anytime (CET evenings but next two weeks I'm in India and Taipei)

In an OGS game against a 3k AI sensei found 22 occasions where I could have played considerably better (mistakes). This is an order of magnitude more than the mistakes I made in the 5H ganes I played against it. I realize that there's a causal effect: if my opponent leaves a major opportunity and I miss it, that counts as a major mistake. AI Sensei won't leave such opportunities so my mistakes are entirely self inflicted. You play better against better opponents because all the mistakes come from your own initiative. Weaker players leave more opportunities to make moves that are worse than the best move, even if they are techically speaking good moves. Against AI, good moves are good moves, period.

In connection with self-inflicted mistakes, I came across this classic quote by the Editors of the British Go Journal No. 55, March 1982, on the deciding game of the Wessex Tournament between Richard Granville and the late T Mark Hall:

This is slightly unfair on Mark, I feel. It seems to imply he often played for a swindle. I’m not claiming he never did that, and of course he was a connoisseur of the pipe tesuji, but his abiding philosophy on the board was “mistakes punish themselves”. That is a remark worthy of intense study, especially by the sente warriors. Worthy of Lao Zi in fact, because it’s what led him to grasp the meaning behind Go Seigen’s notorious comment “White 8 is bad” and further led ys to discover that White 8 is bad in an unusually high number of gamed (at least according to commentaries). That too belongs in a Lao Zi sort of context!