Ohashi Hirofumi 6p plays Golaxy on 17x17

[Uberdude]

https://home.yikeweiqi.com/?fbclid=IwAR ... 1/10169988

Komi was decided by auction, I don't know what it was though this sgf I extracted from the above page says -4.5 (did Ohashi have to play silly 2nd line opening moves in return, or he just wanted to?). Golaxy won by resign. Being able to play on different board sizes and with different komis (unlike bots like AlphaGo-/Leela- Zero whose neural network is trained on a fixed board size and komi) is a touted feature of Golaxy so it's nice to see it in action.

Here's the game, though eidogo can't do 17x17 so imagine the right and bottom 2 lines aren't there.

[Yakago]

I saw on the Leela Zero GitHub (with respect to another Golaxy 17x17 game) that the first moves of the game were played at random, after which there was a komi auction. In that game Golaxy furthermore received another 10 point handicap, but lost the game.

Perhaps this game has similar rules in place

[macelee]

This follows a special rule called 天弈 in which 4 stones are randomly placed on the board before the two players each offering a komi to take black. Obviously the komi would be very large if the existing stones are in very bad positions. After the biding process, Golaxy will give an additional 10 points to make these games more balanced.

Golaxy is probably the only strong AI that can handle arbitrary komi and board size without retraining.

This is seen as one possible way forward for online Go - with this 'random' setting it is much harder for people to cheat with AI in online games.

BTW, Golaxy has recently released training software (in beta) among a small group of Go fans. There is a playing mode and a problem solving mode. In the playing mode, Golaxy is said to play intuitively therefore using very little computing resource. It is still incredibly strong. I have only managed to win one game in its intermediate level and have no chance in its advanced level (there's another level above advanced).