Maciej Celuch vs. Leela Hex, Havannah
81 replies. Last post: 2019-04-30
Reply to this topic Return to forumReturn to forum
You have 0 new messages
You have 0 games on move.
You have 0 invitations to game.
81 replies. Last post: 2019-04-30
Reply to this topic Return to forumCan anyone contact Maciej outside of LG to encourage him to sign up for the about to start Hex Championship?
With or witout Maciej, I think the strongest Hex Championship ever has started.
I say, we should set up a fund, and the first HUMAN to win a tournament will get the prize! :) :)
If leela or gzero win this tournement, then the prize is conserved for next tournament, and so on.
Leela messed with my head so much that I cannot even find a good opening move anymore
Arek, I'm also unsure what to play as first stone vs shalev. Suggestions? What are the fair choices here? :D
I downloaded leela and gzero's games to have a look at their opening preferences. They are both pretty varied until at some point in November, where leela started playing consistently: first c2 every game, then more recently a13 every game. Leela swaps a13 if you play it against her, but does not swap c2.
Interestingly, the bots seem to hate most openings on the 'a' column. You can sometimes catch them playing a3 or a4 (they swap a4 but not a3), as well as a10, but never a5, a6, a7, a8, or a9. This is despite the fact that the bots experiment with crazy moves like l3, f3, h3, and (the weirdest one) f2, a nonsense move that leela played several times as recently as November. Gzero sometimes plays k2 but leela never does.
It's unclear whether the bots swap most stones on the 'a' column, because we humans rarely play them either! A while back leela used to *not* swap a5, a6, a7, a8, and a9, but this probably changed recently (the only recent game I can find is this one where leela did swap a7).
I think I will open my games with a10. It's been a while since we've seen it in top level play, even though the equivalent of a10 was the hardest stone to solve on 9x9, and the bots agree that a10 is reasonably balanced.
shalev, make a serious effort to win against leela! If you need tips you can ask here, we'll help, heheheh. All humans vs the bot. I'm thinking about 6 g7 in my game.
shalev, so you're also rather clueless about A column as first stone. Same here. I'm not even sure a10 is weaker than a9 or a8 or a7.
I3, F3, G3, instead, they're reasonable. We tried the corresponding on 11x11 and we've found interesting games. On 11x11 they still win but on 13, who knows.
Lazyplayer, for your game, 6.g7 looks good but be careful not to delay the joseki too long. How do you plan on attacking the d5 stone? If you're sure you're going to play d4 later, then you might as well play it now. If you're keeping your options open for something like e4, then I guess it can make sense to delay the joseki a little bit.
Ehehe 7 b6 is something crazy that I invented in a blitz game against Maciej on playok. I think it's crazy.
For my game, my feeling is that the key is finding a refutation of black L6 and black L3 idea.
And yes D4 is an obvious choice but I don't want to give away possible future choices. There are a couple more ways to attack top-left for white.
6 h5 is the other obvious option in my game. Hmm. 6 g7 is rather crazy to be honest. Beautiful from a purely geometrical point but is it perfect play…
6 f9 is the one I like more. Or more precisely, the one I dislike less. In fact I tend dislike every move because I always see plenty of flaws and opportunities for opponent to kill me. :)
Worst thing is that the leela-gzero game is total nonsense in my eyes. What were they both doing?
Leela the 4-4 maniac even played 12.f5 to compliment it's 4-4 haha. Later brilliant 18.g5 hahaha. And to conclude 36.e13 to the rescue!
WHAAAT? :D
Arek, gzero has played a mess, it's its style, leela is more ordered. And I guess leela must have made an error early in the game.
Basically, I think it's just luck, and when playing against leela, everyone need some luck.
Maciej should show up again in January, when vacation days are restored. At least, I hope so. He needs to play TwixtBot also :-)
@lazy maybe if I refactor the code to not be so much of a mess, it will stop playing a mess! :)
Caroll, j10 or i10 must be move loss. You can also say that move 2 is the loss and it should have swapped, but I don't agree with this.
I've not analyzed in depth eh, I'm just telling you my gut feeling about that game.
If leela wins, then it's wow for spectacular win. If Daniel wins, then wow for a finding a reasonably simple win. :)
Yeah, I've been looking at it. Very curious to see if leela can pull a rabbit out of that hat, her position looks terrible.
Actually, 7. f7 looks so bad that it almost makes me suspect leela is buggy or something…
Ehehe, 7 f7 maybe is genial or maybe it's garbage, we'll see shortly. After white 20 f9, then come 21 e9, 22 e8, 23 d9! This is the trick that may save the game.
I see what you mean, but it doesn't feel like enough if Daniel plays it right. I could be wrong, we'll see (I don't want to comment in detail in case Daniel doesn't want us to)
shalev, we can discuss the variation that has been discarded by leela. what would you do here? I think without this variation, leela was lost.
Well, daniel just resigned. My guess is that he blundered with 24 i6. I'm unsure if leela could 21 e9 can do something for leela.
Maybe leela can share its own analysis? Maybe leela isn't that strong tactically, but it has good sense of board-wide intuition. :)
And also wow at this, I had missed g1 here: https://littlegolem.net/jsp/game/game.jsp?gid=2036019&nmove=61
I think even the bots can miss these “hex brilliances”, for whatever they're worth! :D
Game #2036019 confused both the bots, they were both intent on fighting out the RHS battle (which was trivial) before diving into the LHS one. After https://littlegolem.net/jsp/game/game.jsp?gid=2036019&nmove=55, then gzero searched out the LHS and jumped from 60% probability to 95+%.
Incidentally, and FWIW, I updated gzero to a brand new model today (with almost no testing)… and reduced search time to 5 minutes because (a) I don't think it really makes any difference in 99% of the moves it makes, and (b) it is using too much of my computer's memory.
No, doesnt work indeed… Dont you dare saying leela is not that good tactically haha
Arek, leela played this, and it seems a white win to me. I'm unsure about this because I had no time to analyze it.
Maybe I'm missing something, but I see only this continuation or basically equivalent variants.
Yes, seems like game against leela was mine, but one blunder and I lost. Another intresting game vs gzero
Daniel, yes, well played, I also see my variation is worthless. It's surprising to me that leela abandoned d5 and allowed you to win so directly…
Daniel, stupid blunder by gzero, he has changed the code without testing at all…
Well, you had bad luck with leela, and good luck with gzero, you get a fair result.
gzero was losing all along. I thing gzero said that his change probably will have no effect. Congrats Daniel on 2 strong games vs bots
Daniel, oh sorry, I just thought there was another column on the right, damn I'm stupid. :D
Bill, but who is winning? My preferred hex style is all or nothing, no middle ground and no moderation. :-)
Bill, my game with Daniel is also rather similar, so all my chances in this tournament all depend on the same game.
Bill, my game with leela is almost over. Honestly I'm playing fast because I'm curious to see the result. But I'm not optimist. It should be leela win, but I'm not sure. Best hopes for saving the tournement now are on Arek. I'll try to win the other games so I get a decent standing at least. :)
Losing to bots by copying from bots… What a sad story. Never in my life I have checked a position more thoroughly and the conclusion is that the following position is losing (white is losing)
http://www.trmph.com/hex/board#13,k12i5d5d10k4i7l9f9
Personally I will avoid playing 4-4 again, but I'm happy to hear suggestions for move #8 other than f9.
Assuming black default moves maybe this is hopeful is it?
Nah, my suggestion is not possible. All white moves lose to black e9 like this
How do you see your chances against leela and gzero? Are human still ahead in hex? :P
Force, probably if humans really thought hours for each game, and really played a lot of high quality games, then they would be able to compete.
Basically there are no professional Hex players like in Go. A professional Go player is MUCH stronger than an amateur playing casual games.
Sure, no doubt about that. It's a shame that such an profound game is so unpopular compared to Go, chess or even draughts…
IMO it's neither an inherent shame nor a benefit that Hex is so obscure. It's just the way it is. No one is likely to throw much money at it, but computers can investigate ahead of us. They may even become excellent teachers, once they get past this human language stuff.
Force, at least on 13x13 leela is easily the strongest player at the moment. It's actually the perfect level to give fun to humans, because it's beatable and it's a really cool challenge to do it.
Strongest? She is still 170 points shy of Maciej …
which may mean she is still not that good or that they haven't played enough games to show their true (relative) strength.
And to add to what Arek is saying, if Leela goes first, the human is always starting with a theoretically won game due to the swap rule in that situation no matter how strong Leela is :)
The question is wheter Maciej is still the best human player, he gained his points quite a while ago. I would love to see him competing again…
Any information, whether Maciej's going to participate in next championship?