Zen team to hold press conference November 9

[yoyoma]

From Aja Huang's Facebook https://www.facebook.com/aja.huang/post ... 9867940581

The news says the Zen team will hold a press conference in the Japanese Go Association on the 9th of November. Also, several Chinese Go programs have reached professional level. I'm glad to see the AlphaGo Nature paper has made significant contributions to the computer Go area.

Japanese source: http://www.yikeweiqi.com/bleachery/22444

Since Zen19K2 reached 10d on KGS and #2 on the top 100 list, we haven't seen Zen on KGS. But several versions have played on CGOS, most of them running smaller hardware (1 CPU, 0 GPUs). Version 12.4 which played on KGS is rated 3606 on CGOS, 12.6 is rated 3670. Not a huge gain but maybe the gains are larger with a GPU? So far only 12.5 played with 1 CPU and 1 GPU, reaching 3846(!)

Let the conjectures begin on what the Zen team will be announcing!

[Bonobo]

Japanese source: http://www.yikeweiqi.com/bleachery/22444

Chinese source, I’d say, not only from the “weiqi” in the URL, but also from the script (which I cannot read; but I used Google Translate to verify )

[mschribr]

First, I want to see Zen play Ke Jie. AlphaGo has been gaining 1 stone a month according to Deepmind. I think Alphago is too strong for Zen. After Zen beats Ke Jie then Zen should play Alphago.

[pookpooi]

AlphaGo has been gaining 1 stone a month according to Deepmind.

Actually the DeepMind presented it as 'one rank per month' (AlphaGo gaining slower when reach professional dan strength).

[Mike Novack]

Except that "single machine" is deceptive when applied to one with 48 CPU's and 8 GPU's

[pookpooi]

Except that "single machine" is deceptive when applied to one with 48 CPU's and 8 GPU's

You make me laugh a little bit because it's so true!
Even the DeepZenGo project which is sponsored by Dwango's Kurisu server has just 2x quadcores CPU and 4 GPU.
But the 'distributed version' is more deceptive to me, it sounds too simple to be the machine that Google throw everything inside it.

[Uberdude]

AlphaGo has been gaining 1 stone a month according to Deepmind.

Source? I've not heard anything from DeepMind about progress in the last few months (just release of commentaries).

[pookpooi]

AlphaGo has been gaining 1 stone a month according to Deepmind.

Source? I've not heard anything from DeepMind about progress in the last few months (just release of commentaries).

I think that he misquote from this

[mschribr]

AlphaGo has been gaining 1 stone a month according to Deepmind.

Source? I've not heard anything from DeepMind about progress in the last few months (just release of commentaries).

I think that he misquote from this

Correct pookpooi. I meant to say 1 rank per month. Aja Huang and Fan Hui spoke at the US Go Congress. Huang said 1 rank per month. See the video Opening Keynote: "The Story of Alpha Go", presented by Aja Huang 7d and Fan Hui 2p. The graph of 1 rank per month is about 1 hour into the video. The video is at https://youtu.be/LX8Knl0g0LE?t=3624

[Uberdude]

So they said one rank a month until March 2016, extrapolating that continued rate of improvement to now seems a little optimistic without any evidence...

I know a month or two after the Lee Sedol match, I forget exactly when, there was that talk about new AlphaGo can give old AlphaGo 4 stones or something; but such self-play handicap games are rather unreliable at measuring strength differences (I'm not doubting it has got stronger, just how much). As far as I know AlphaGo hasn't beaten Lee Sedol in a new secret match on 3 stones.

[mitsun]

It is always fun to speculate how far we are from perfection. I have heard pros express confidence that they could beat God with something like 3-4 stones. So Alpha Go cannot continue to improve one stone per month for very much longer

[dfan]

So they said one rank a month until March 2016, extrapolating that continued rate of improvement to now seems a little optimistic without any evidence...

I know a month or two after the Lee Sedol match, I forget exactly when, there was that talk about new AlphaGo can give old AlphaGo 4 stones or something

That was a misinterpretation too.

[yoyoma]

I'm hoping they will be announcing a challenge match with a strong pro. The Zen team is Japanese so I would expect a Japanese opponent... Iyama Yuta?

The best point is Zen is a commercial program. They should be motivated to work on improving the program when running on regular home PC systems and selling that to all of us so we can benefit.

Now to stir things up a bit: If Zen beats Iyama Yuta, obviously some people will be quick to point out maybe he isn't the best in the world Of course I'm talking about AlphaGo! Imagine Zen challenges and beats AlphaGo. A nice way for the Deepmind team to exit a field that they don't seem keen to stay in, and for us to get some closure on that front. Is it possible that Zen is stronger than AlphaGo? Seems unlikely, but what do you think?

[pookpooi]

The best point is Zen is a commercial program. They should be motivated to work on improving the program when running on regular home PC systems and selling that to all of us so we can benefit.

Actually Zen is also very best at developing to run on home PC. Do you remember when Zenith Go 6 was released back in June?
Right now Zen running on a single core is one stone stronger than that version running on dual core.

[Jeromin]

Right now Zen running on a single core is one stone stronger than that version running on dual core.

Single processor, surely. Are there any single core pc for sale anymore?