no room for errors

A game versus leela taught me a useful and a very unusual lesson.

Actually I've learned from leela's bad play this time.

I argue this position is black win https://littlegolem.net/jsp/game/game.jsp?gid=2005726&nmove=14

But there is a very tricky dilemma there. Black has to decide between two options:

In the first one black has solid connection to bottom which is very tempting. So tempting that I didn't take time to consider other options long enough. Unfortunately it possibly is a losing line.

Second one is tricky because it loses a direct connection to bottom. Despite that, I argue it's correct and also winning for black. It seems so that undermining the white central group is way more important than the connection to bottom.

I distrust leela's ladder on bottom. IMO it's nonsense, however, my ignorance led to leela profiting from the nonsense and finally beating me up.

Arek, in the other thread, I was arguing that "being connected is not enough". This game by leela was just the most outstanding example. :)

Another option that was maybe worth analyzing was this: http://www.trmph.com/hex/board#13,m1m1j9d10d5f9b11b12c11c12d11d12f11e11f10

This is such a mess, but you know I love mess... :D

Your proposal is also surely interesting. I think it would have been more principled to not play b11 at all.

Also, like I said in the other thread, leela probably makes up for its weakness in understanding with extreme consistency.

Is leela a good hex ai? i am a bit out of the loop regarding those.

I favor the foldback option

Christian, yes, leela is very good. Probably 2500 or so, maybe higher.

Alternatively, play as in the game but don't follow up with 25.k3. Is white winning here? I think I prefer black there...

Shalev, your "foldback option" looks very good indeed. Your L7 idea, I would take white.

But even the "foldback option" is far from a sure white win.

I mean, far from a sure black win. White still has hopes.

Well the foldback is the simplest and actually almost identical to my proposal.

The difference is in when white should tenuki, I argue after foldback white tenuki now

however in my proposal it's either early or the same, but worse position than foldback

Arek, this is nonsense. In the foldback variation, white should tenuki here.

True. Anyway, I'm going back for what was mine :P

https://littlegolem.net/jsp/game/game.jsp?gid=2015384

Arek, you convinced me that your proposed move (15. g10) is pretty good. I didn't see the point at first, but I see it now.

Arek, leela changed his mind? Maybe it's a new version, it's already stronger! Damn! I had hoped to replay same position in the next tournement.

shalev, but your proposal and his proposal are good, sadly, he didn't play neither in the real game!

Please replace my "but" with "both" ;)

Arek, shalev, lazy: thanks for your insight. I'm astonished that despite being certain that leela played nonsense move, it is still very hard to punish such mistake. What do you think is its strength?

Force, consistency, even when it is wrong, it is very close to being right anyway. The human player will make a small error and then it's winning.

More formally, my argument is that the errors are "small", that is, they can't be easily "punished".

Something similar is true in other games, even chess. Even chess engines make errors, but you've to compute next 30 exact half moves to prove them wrong. This is not practical so you'll simply end up making another error and the game is again even. If you make an error however they'll probably spot it and punish you. This is why the human players lose.

???

The next Hex tournament will be something to watch. Do we think the end is in sight for human champions?

Bill, hehe, maybe my answer isn't saying anything at all. Yes obviously it wins because it makes fewer errors. The more interesting question I guess which kind of errors it still makes? Well, hard to say. Probably it misses the sequences that require more experience. But then, compared to humans, probably it's still stronger even in these cases.

Bill, me and Arek will make a serious attempt at next tournement to prevent that. I'm sure Shalev and Daniel will also make their attempts. We'll see who wins. It's even possible that someone wins against the bot but then loses against other human players and the bot will win the tournement anyway. Ahahah. We'll see! Do you want to encourage us to play better? Then set up a prize! :)

Should be very exciting :-)

In my another game leela changed it's play but I did not! I'm experimenting so that I have more knowledge about leela during the championship game :]

Arek, it's game over already, leela loves quick wins. :)

Perhaps we can get MacieJ to play as well :)

Is anyone in contact with Maciej ?

Perhaps Its the last chance to beat leela :/

Maciej quit playing because there were no opponents at his level, I was still so weak, Arek was not much better, shalev was not here...

And by the way, we can learn and improve from leela play, so it's not bad news actually. It's just that we have to reach perfect play to have any chance of winning vs leela.

I would like to see some game of leela vs itself on 19x19... need to train for that and probably it's so much harder.

Does leela play at size 19? Or is she strictly limited to 13?

Lazy, don't you think that now there may be few players, that could give Maciej some decent rivalry? Isn't he curious about leela's play?

I support leela and thinkg that it's good source of knowledge, however somehow I don't like the idea of computer program becomming a champion.

Btw is leela’s level fixed, or is it still improving?

Still improving. Maciej probably hasn't been informed there is now a worthy opponent.

Leela will improve forever because it's using same code that is used for Go and Go is always under development.

Leela's code is likely in almost fixed state and that it is the network that is mostly improving from further self-play/training.  Only Leela's (hex) developer can say for sure how it has improved so much in the last few months.  There were large changes in the code base between leela go -> leela chess, so i assume there also largish changes to make it work with hex also.

What about gzero_bot, is it same base and is your network still evolving, what's its size... ?

@Carroll, gzero_bot current network was trained pure supervised with quite a large network (~128 filters, 8 layers... guessing, I can't remember) from the self play generated up until July (when we had bot competition).  I attempted to add in all historical LG games on top of that, but it became significantly worse, and then I tried more self play again and it got worse also.  :(  Not sure what is going on, but I was thinking in starting a fresh run over xmas (ie from zero data).

gzero is mostly an alpha zero clone, but a completely different code base from leela.

Leela do You plan to train for 19x19?

???

Please encourage Maciej to play, new championship will start soon :)