Hiyo~ Recently I started making fairly substantial edits to Sensei's Library pages, in large because many of them are extremely outdated (since before the AI era), and our thinking about many joseki and other positions has evolved substantially.

Since I am only 4k (OGS) and never really studied Go formally, would anyone be willing to check over my edits to check if they are accurate or a fair representation/overview of the position? I sort of believe that the Sensei's Library joseki pages are targeted most to ddk and sdk players (people trying to study without spending a lot of money buying books or lessons), so I hope we can digest information to present the big picture.

Here are some pages that I've updated:

• 4-4 point Josekis
• 4-4 point low approach
• 4-4 point low approach low extension
• 4-4 point low approach low extension, contact
• 4-4 Point Low Approach Low Extension, Slide
• 4-4 point low approach low extension, slide, pincer (partial)
• 4-4 point low approach low extension, slide, 3-3
• 4-4 Point Traditional Slide Joseki (working on this)
• 4-4 Point Traditional Slide Joseki, 2-8 Invasion
• 3-4 point Josekis
• 3-4 point high approach
• 3-4 point, high approach, inside contact
• 3-4 Point High Approach, Inside Contact, Hane
• 3-4 Point High Approach Attach-Drawback Joseki (working on this)

It would even be more amazing if you're willing to help contribute edits/articles/organization to this initiative! Sensei's Library is a wiki, so anyone is allowed to edit it. In fact, I would be extremely happy if people went and started editing pages immediately.

If the wiki interface is too difficult to work with, you can write things in this thread and I'll try to transfer it over to the wiki somehow (although I'm only one person).

Collating threads / forums posts / reddit discussions / youtube videos on a particular subject are particularly valuable. If you remember a discussion or video about a subject, please link it, and I think people would find it useful if Sensei's Library was an index to those discussions. People can also summarize the discussion from multiple places and present it on the wiki, and doing this is totally great even if you're afraid that that you're not knowledge enough on a particular subject. Don't be afraid to be wrong -- Sensei's Library is a wiki and other people will correct it (or debate it) -- and we can all learn from something~!

I'm interested and will be around. I've done spot changes myself on certain joseki of my interest.

Awesome!! Feel free to go ahead and just start making changes!

Ah, also if there are particular pages / topics that people would like to be modernized first, please speak up -- and we can probably find ways to work on those articles.

Great to see a thread on this!

I've mainly been working on the people side of things:

- adding new pros
- updating new pros' ranks and when they achieved them
- adding / replacing profile pics
- adding kanji / hanzi to names
- archiving links
- adding new tournaments
- adding the results of new editions of tournaments

Sensei's is probably more active right now than at any time in the last five years
We even revived the Article of the Week which had been dormant since 2014.

I'm now an OGS 3k, btw, due to the rankbump. I guess I need to update my own rank ^^

Personally, I am in favor of maintaining the pages for "outdated" joseki; maybe you could make a new page for the AI version of a particular pattern.

Senseis library serves various purposes, and I find value in having the old explanations of patterns that may not be AI-approved. For a page explaining a pro's thoughts on a joseki, I'm not really a fan of seeing the win rates, etc., thrown up there.

If I want to know what an AI thinks, I can open up an AI and give it the position I'm interested in analyzing.

The value from senseis library, imo, is different; you get a human perspective, as well as historical context behind why pros thought certain moves were good and bad. Maybe the page is inaccurate or gives suboptimal moves. But I'm not looking for an AI analysis when I'm going to senseis.

Basically, I think there's some element that's lost when updating these pages and just copying the stats you get from your GPU. We don't need senseis library for this type of analysis.

---

I know some people disagree with my opinion, and want to run KataGo on whatever position that's documented on the internet.. That's why I'd recommend making separate pages; AI josekis for those interested in the AI analysis, and the old ones for those interested in the historical context. There's at least some people interested in the latter (me, for example).

_________________
be immersed

Yayy!



Thanks for sharing your insights! And I definitely don't intend to remove any older content or joseki! I've mostly been shifting content and around and pushing things onto different pages (like creating a dedicated page for the more traditional slide joseki) and copy-pasting all of the old discussion there. It is still accessible by clicking on the link for the slide joseki.

My intention with the winning rate shown was to emphasize that actually most of these joseki do not differ much in winning rate, and I was hoping to communicate to beginners that the winning rate difference really is not that significant. However, if people don't like seeing it, I can also remove it, although I think it's important to emphasize to people trying to learn joseki that the AI joseki is often not that much "better" than the traditional joseki and I think the point difference is an easy way to see that.

For me, Sensei's Library was most useful for me when I was a ddk and I was struggling to learn joseki and understand why they are played. The OGS joseki explorer is great, but it does not have explanations. Also, at that level I did not have KataGo or any form of AI running on my computer, so in some senses I suspect that many ddk's aren't reviewing their games with AI -- and for that matter, arguably they shouldn't be.

I like win rate in doing private analysis on games, but I don't like using it as any sort of authoritative source, since win rate for a position can vary:
* What engine are you using?
* How may blocks in the network?
* How fast is your GPU? How long have you let the AI calculate moves?
* What is komi set to, and what is the ruleset?
* How well does the computer's win rate correlate to the realistic win rate for a human? (e.g. something with a lower win rate may be more playable for humans)

It's not that I don't think that the win rates are useful. It's just that I think of win rates in a more fluid sense, in that the exact numbers can vary based on a number of variables. As such, I typically prefer to investigate win rates interactively with software, rather than assuming a static value that was retrieved from one particular version of one particular AI based on one particular number of play outs.

That being said, I may have an old fashioned view on this. Or maybe I'm simply missing the pre-AI era

_________________
be immersed

Ah, also I finished updating/writing this page: 4-4 point low approach low extension, contact

This was mostly original research looking through waltheri and trying to guess why professionals played one variation versus another, so I would appreciate feedback and corrections.


>__< I actually totally agree! I had a footnote every time I placed a win rate hoping to communicate that the numbers aren't authoritative and can change widely for a huge number of factors. But if it's being overlooked or misinterpreted... yeah, I guess that's a problem.

I'm trying to figure out what would be the best way to convey to beginners and ddk's browsing SL which joseki are "good for Black" or "good for White" -- i.e. most people would agree that the 4-4 point one-space high approach is good for the defender.

Maybe it would be best to just state it verbally in words?

But the difference is actually somewhat small -- the magnitude of how much something is good or bad is difficult to convey. At least at the ddk level, even remembering a joseki correctly can be considered success.

Your view is the good view. Viewing the winrate as an in-and-of-itself meaningful number without reference to context is from the early-AI era when people didn't know how to interpret them, but I think understanding of winrates has spread a bit more.

Understanding of score hasn't spread as much though. People verbally talk about a score prediction of +2 as "AI says Black is 2 points ahead" rather than "if Black were to give away 2 points, AI would be indifferent/uncertain which side it prefers". I'm sure many people mentually understand the latter but say the former merely as verbal shorthand, but I'm also pretty sure that for many people the shorthand is *also* what they mentally think, or that they are confused about what to think.

I removed the scores for now, since it seems like a controversial subject, and it's unclear how helpful it really is to show them! o7

Not out of stubbornness but I don't see the difference between the two statements, or at least not between the meaning of either statement pre/post AI. Before AI, when pros said "Black is 2 points ahead", except for the late endgame, what else could they mean than "if Black were to give away 2 points, I (pro) would be indifferent/uncertain which side I prefer"?

I think there's a difference between

• If both sides keep trying to win the game, Black will be ahead by two points at the end, and
• If White is gifted two points, then both sides keep trying to win the game, it will end in a tie,

because play may be different in the two cases. This is probably more clear if we imagine a much bigger lead, like twenty points.

I'm imagining a position where there's a potentially dead white group on the board worth, say, 50 points. Let's say there's 50% certainty that the group lives.

Even if the group dies, the game is close - but black is leading by 2 points at the end of the game if the game ends and that big white group is still dead. So maybe in this situation, we can say that "if Black were to give away 2 points, AI would be indifferent/uncertain which side it prefers" - because the result of the game would depend on whether that big 50 point group lives or dies. The end result may be W+48, though. Or it may be B+50.

Comparing to another extreme... Let's imagine the same situation, same 50 point uncertain group. But this time, black is leading by 49 points at the end of the game if the white group is still dead. So in this case, I guess black could give at least 49 points to white, and we could still say that "if Black were to give away 49 points, AI would be indifferent/uncertain which side it prefers". The end result may be B+49, or it may be W+1, I guess - based on whether the white group lives or dies.

Not sure if I understand this completely correctly, but maybe in these cases, we know the game will not be a tie when gifting white with X points. Is it wrong to say that "Black is 2 (or 49) points ahead"?

_________________
be immersed

I'm certainly not knowledgeable to any agree (and I'm probably wrong), but I get the impression that the AI doesn't believe in life-death uncertainty in that way, in the sense that it believes groups are alive if it thinks its alive or dead if thinks it's dead. AI are just that much better at calculating endgame than human professionals, for that matter.

A 50% uncertainty that a group is alive = miai for life (kind of) (or one side has objectively more ko threats).

Or well, there are no tsumego problems with an "uncertain" answer, in a sense.

The fluctuation in score, at least to me seem to largely happen when both players repeatedly miss the vital point for their giant dragons. The vital point might be obvious to AI, but not necessarily obvious to humans. The AI score in this since isn't a prediction of whether it thinks Black (human) or White (human) would win -- but rather a score of whom it favors assuming that AI took over and started playing itself Black (AI) vs. White (AI) at that precise moment.

I don't think AI sees the concept of life and death the same way as humans, either. Here, I'm more interested in trying to explain the difference between saying that "Black will be ahead by 2 points at the end of the game" and "if white is gifted 2 points, it will end in a tie". I don't think it matters if we are talking about AI or not here.

But even in the case of AI, in theory, I believe there should exist board positions in which 50% of play outs end up with a group dying, and 50% of play outs end up with a group not dying; you'd just need to make the board complex enough. Sure, humans make a lot more mistakes than AI and the life/death status of groups may unintentionally fluctuate a lot. But AI is just making estimates based on imperfect play outs, too.

_________________
be immersed

I believe the statement Lightvector infers is still different:

"If Black gives 2 points to white, then the probability of winning is 50-50"

Except for the late endgame, neither AI today or pros before, were claiming the game would result in a 2 point victory given sharp play by both. Maybe 2 points on average but even that. I really think pros meant "I prefer Black, and if you give White two points, I'm indifferent to take Black or White" which seems to be what AI "says" when marking a 2 point lead for Black.

The two statements:
"AI says Black is about 2 points ahead"
"AI would be be unsure which side it prefers if Black were given a penalty of about 2 points"

...can mean the same thing if you interpret them one way, but they can also mean different things with a different interpretation. The whole point of the second phrasing is to make it easy for people to gravitate to the most accurate interpretation. (Let's brush under the rug for the moment that due to how MCTS works, even the second statement isn't exactly accurate, such as getting winrates and scores with mismatching signs. The second statement is a good enough approximation for now).

The first statement sounds like it is a claim of an objective fact about Black's lead. But in what sense? Is it a claim about the game-theoretic optimal value of the position? Not really, in general we have no idea what the optimal value of an arbitrary 19x19 game position is, and likely AI is far enough from optimal that even if we knew it wouldn't necessarily even be useful. For example, it's quite plausible that there are positions where two equally-matched top bots would win more as White, achieve a positive average score as White, prefer White (at their levels), but where the game-theoretic optimal score is positive for Black. The absoluteness of the statement also makes it easy for the listener to forget this fact that different players may value a position differently, and leaves ambiguous with respect to what standard the score is being measured.

The absoluteness of the statement also immediately leads to the question of how certain that statement is. What is the chance it is "wrong", or what is the range of uncertainty? It leads one to also wonder why bots mostly don't provide meaningful confidence ranges on these scores, and hides the fact that asking for confidence ranges in the first place is sort of a category error (and the natural question being a category error and thereby nonsensical is partly why they are hard to provide).

The second statement is much harder to misinterpret:

* It's immediately more clear that "2 points" is not an objective claim about the position itself, it's a claim about the bot's subjective preference in that position if 2 points were lost.

* It's immediately more clear what it's not: It's not the game-theoretic value. It might not be what average score would result if you *actually* took the bot and played a million full games with itself from that position (the bot's preference may or may not not match up with this kind of rollout). It's not necessarily how you should value the position, but it might be, you can be smart about it (e.g. are you talking about pro-level play or amateur-level play from that position? Is that score predicated on an inhumanly precise sequence of play or does anything work? Does the bot seem to have any blind spots in upcoming tactics, etc.).

* It is clearer now why asking for a confidence range on that particular specific score output "Black +2" is sort of a category error. How likely is it to be correct? It's correct. This is the number that reflects that bot's preference when it was run with those settings on that hardware for that amount of time with that random seed. It's correct because it was intended to be a measure of the bot's preference in that instance, and it was indeed a measure of the bot's preference in that instance. What is the confidence range on it? 0, I guess. Either that, or else the question is categorically nonsensical. Or else if we want to talk about average error with respect to something else (e.g. the average score difference that a random 1 dan human amateur would achieve a 50% win chance at, or perhaps the amount such a player would win by on average, or the bot's own new judgment after 10 self-play turns), we need to specify what that something else is.

I hope that made sense. I guess this is all off topic though for the thread.

I want to say that I appreciate that you take the time to give clear explanations like this, lightvector. I also get the sense that you are pretty precise with the phrasing you use - something I could learn from.

_________________
be immersed

Another page updated - The first move of the 4-4 slide joseki - 4-4 Point Low Approach Low Extension, Slide

I think my focus for now is on DDK joseki.

I bumped all of the old forum-style discussion onto a "Discussion" section of the page (or moved it to the most relevant subpage where the move is discussed).

I think I also realized that SL is probably most effective for providing real-board examples, which isn't present in most joseki dictionaries. The idea for now is to select professional game where a less popular joseki variation is played and explain why the professional played that particular variation.