Create your own.
Opinions! We all have them. Find out what people really think with polls and surveys from Crowdsignal.
closed
Opinions! We all have them. Find out what people really think with polls and surveys from Crowdsignal.
I don't know what rules are when running on KGS, but if there will be ~2min in recalculation on move before overtime, then MoGo v4.86 with patterns on >|=16-core computer with >|=64GB of memory (for such number of cores and time) will win, I think, with ~8:2...
I couldn't get Zen (and MoGoTW and Zengg) for testing, but as I heard, they aren't weaker...
To pookpool: As I live and earn in Japan, with 15 years of no inflation, the value of $1000 didn't change so much for me ;-)
To Q30: Zen, as running on KGS, appears to be strong enough to give John a run for his money. I don't believe Mogo would, and I'm confident Fuego would lose even if given 16 cores.
Hopefully we can set up a return challenge with John this December, perhaps playing Zen on KGS.
Luckily, 1000$ in this day is less value than 1000$ the day you made bet.
If there was "professional" time limits (i.e. ~2min in recalculation on move before overtime) and >|=16-core modern hardware, John lost all games (with the big degree of probability) to:
MoGo (and furthermore MoGoTW);
Zen (and furthermore Zengg);
and probably Fuego.
SHORTEN THE TIME!
John Tromp is european 2d, best programs are near 1k, so it is very likely that the bot will win 1 or 2 games... Then everything can happen as John is not a machine...
To Mark Schreiber: yes, I agree. I think this match will tell us more about the level of computer go than the handicap games that get played against pros. Though I appreciate that "Beat A Pro" makes a more impressive headline.
I think if John played today, he would win because John is European 2 Dan. The strongest program is KGS 2 Dan. However, in 6 months, the programs will improve and it should be a close match. There should be more matches with computers and amateurs at the same level without handicaps. Good luck.
Today I played against Zen. He still has some problems estimating the value of moves in fuseki (he plays tenuki during only-move sequences), but as soon as this bugs are removed, I think Zen will be an average EGF 2d. He is already kgs 2d and wins more than 50% of his games - without his 20k fuseki moves i´m sure he´ll reach kgs 3d soon.
Regarding "a more relevant test would be multiple game", it will be best of 9. Also John follows computer go and will be familiar with playing style, strengths and weaknesses of the strong programs.
London Open in December 2010 may actually be possible, but with me in Tokyo and John in New York some or all of the games may have to be played online.
But maybe we can find a sponsor.
I'll bet on John but a more relevant test would be multiple games. As with playing a computer in chess, it can be hard at first and then - like the good piece of wetware that we are - humans sense weaknesses and can "game" the program (at lower levels of course). I hope John has a chance to learn his cyber opponent's flaws.
But he better be ahead before the endgame.
When is the match? The London open in December 2010?
And PS, will the game use Tromp/Taylor rules? ;-)
It's too close bet,I think.
MCTS based programs I have ever seen is good for leading position, but in losing position,the programs choices too exploit play. I think that, To maximize winning percentage is not good tactics at losing position. I think programs plays at the losing position, allow to keep a small lead for opponent, and wait a chance for reversal.
It's a close bet, but I think somebody will be able to get a computer with 32-128 processors to play European 2d by the end of 2011. There are multiple teams working on it, and a great rivalry amongst them. You might even find somebody who can stretch the "can be brought to the site" rule a bit. The big sicortex 5832 computer, for example, is portable, for some definitions of portable.
This bet is still too close to call. I think I know what I have to do to get Many Faces to European 1 Dan, but I don't have much time to work on the engine. I also don't know what computer power will be needed. I don;t think any of the current MCTS programs are capable of European 1 dan today, no matter how much computer power they have. There are some algorithm issues that need to be fixed first. But the algorithms are being actively tuned by several people and teams, so European 1 dan should be reached within a few years.