December AI Updates

When I was looking at the leela zero code I believe there is now an option to train other go boards. But that does take a lot of computing power… oh well, thanks for the consideration.

1 Like

Yep that’s true, though 19x19 took a metric ton of computing power, over two years by the community, 21x21 and 25x25 would take exponentially more, it would be a major undertaking.

2 Likes

What about “ask score” button?
So KataBot will just tell you score during ongoing game
(without drawing territory and chances)

4 Likes

Doing something like this, which is effectively replacing our sub-par score estimation function with a much superior kata go implementation, is something I want to do, but it’s not quite so simple to do it with low enough latency and reliability yet. It’s on the mid-long term agenda to do though I think.

3 Likes

Maybe I will join the KataGo project and try to extend it to larger boards. Would love to see that be a reality, both as a go player and as a programmer.

4 Likes

I would dislike having strong AI provide any estimation about an ongoing game (whether it be score, territory, or even just win probability) to the players of the game.

If the score estimator were to be improved with strong AI, I would hope that the players of the game would not be able to access it (during the game), or that the feature would only be present for unranked games.

I made the following argument in another thread:

On another note, I actually hope that the “ongoing game score estimator” (I only mean the one that is available to players while the game is still ongoing, since using AI as tool to help in stone status and scoring disputes is different and potentially helpful) is not improved with strong AI. Right now, the score estimator is terrible, so I don’t think too many people take it seriously and it doesn’t impact too many games. However, if players had the ability to query a strong AI to evaluate the current position, then I think that would be a form of assistance that should not be allowed.

For example, sometimes my opponent would make a blunder that I would not realize and fail to capitalize on (essentially one blunder followed by another). However, with a strong AI giving score (or even just win probability) estimates, that might tip me off that my opponent’s last move presents an opportunity that I should search for, and thus a decent score estimator could provide unfair AI assistance.

5 Likes

There is option to turn estimation off.
It should exist or not exist. There is no sense in nearly random answer about score.

2 Likes

What komi does KataGo assume for 9 x 9 and 13 x 13? Default for Chinese rules on OGS is 5.5 for each, but I’ve found that most people believe 7.5 would make more sense for each. Anyone know which komi KataGo was trained for?

3 Likes

I believe KataGo is unique in that it handles variable komi. Someone correct me if I’m wrong

3 Likes

It does handle variable komi, and we feed it the komi from the game. As for training, my understanding is it takes the 19x19 network and some some clever hackery to make it work with other sizes, so while it’s probably technically weaker on non 19x19 boards, it seems to do pretty good.

3 Likes

Maybe not the right place but can anyone explain what’s going on here? White won by 0.5 but AI says black wins by 1.5… thanks!

2 Likes

The AI is showing a difference between Chinese and Japanese rules, as you said. The game used Japanese rules; the AI used Chinese rules. If you look back a few moves, you’ll see that the AI wanted Black to take the last dame instead of filling the ko. That’s characteristic of Chinese rules but makes little sense under Japanese rules. The AI also says that White’s losing move was passing instead of taking the last dame. That’s another tell.

P.S.: it would be cool if @anoek could configure KataGo to use lightvector’s territory ruleset in the KataGo config file when it evaluates Japanese ruleset games. But that may not be easy to do.

5 Likes

Wow - that is seriously awesome! :slight_smile:

3 Likes

Thanks, and so Chinese needing 7.5 komi top get the right result? Fascinated to know what the AI graph should look like! Presumably could be quite different with 7.5 komi
Or is AI graph always going to look odd if there’s only half a point in it?

2 Likes

Is there a GUIDE on using the AI? I just noticed it for the 1st time and messed around with it, but I would like to understand how to USE it, and understand it (for example there are numbers for some critical moves that have zero meaning to me atm).

I did check the FAQ quickly but no “AI Guide” is listed, except something not related about setting up custom bots.

3 Likes

Yeah I thought I was, clearly not, I’ll look into it

5 Likes

I have not written a guide, some sort of integrated help for the site in general is something that I’ve been wanting to implement for quite some time now.

4 Likes

Although really you should take the rest of the year off! Thank you so much for all you do!

6 Likes

Thank you for all your efforts. It’s greatly appreciated and I intend one way or another to support OGS one way or another, so the question for the guide on what the AI means in it’s review is of course asked as a “plus” to that. It’s still very usable on variations the AI spots without really understanding the numbers/weights. Thank you again.

3 Likes

Try this:


The screenshots are the old design, but it should be still correct.

6 Likes