AlphaWhat?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.

No comment by Marco Costalba. Why?

AlphaWhat? Never heard of it...
1
3%
Of course, they lost many "private matches" vs. SF + book + proper hashes + Syzygy.
6
15%
This is an abuse of Stockfish name and reputation. I'll start a legal action.
0
No votes
It's far better than Stockfish. Very promising approach. Congratulations!
17
43%
Think at it as you want. That's not my business.
16
40%
 
Total votes: 40

Tobber
Posts: 379
Joined: Fri Sep 28, 2012 3:53 pm
Location: Sweden

Re: AlphaWhat?

Post by Tobber » Thu Dec 28, 2017 8:18 pm

Lyudmil Tsvetkov wrote:
Tobber wrote:
APassionForCriminalJustic wrote:
shrapnel wrote:People get so involved in discussions about ELO, TBs and TimeControls and other technicalities, they forget the basics of chess.
If you just go through the Games as a chess player, its completely obvious that Stockfish was completely out-classed and simply didn't have a clue about what was going on.
It is obvious to any average chess player that AlphaZero was playing at a completely different level and was seeing the Chessboard as a whole in a completely different way than Stockfish.
You can grumble about the version of SF used, TC used and anything else, but in my humble opinion, AlphaZero was so completely superior to Stockfish, that it wouldn't have made any significant difference to the outcome.
Its the Dawn of a New Age in computer chess, whether the nay sayers agree or not.
Of course, unfortunately since DeepMind isn't showing much interest in chess, the voices of these negative people will only grow stronger.
Soon they will claim that there was nothing like AlphaZero and it was all a big hoax.
But the people who really understand what Chess is about, have seen what actually happened on the ChessBoard and know the Truth.
You need to remember that the conditions were not at all favorable to Stockfish. This is not about being negative. It's a fact. Stockfish only had one minute to move. That in and of itself is certainly going to make it weaker. Then you can talk about the hash file size. But to me AlphaZero clearly had a hardware advantage running on 4 TPUs. Yes - I am sure that this point has been made multiple times but people seem to agree that those 4 TPUs are many, many times more powerful versus Stockfish's 64-core rig. Team Google should have had more balls. They should have made Stockfish as absolutely as strong as possible. That way - you cover all of your tracks and ultimately diffuse the inevitable skepticism.

Despite all of this Stockfish still had nearly 80 draws. That is far from your claimed whole new different play of chess. Maybe AlphaZero is the future of computer chess. But not anytime soon... good luck affording the particular hardware used... plus AlphaZero is not really that impressive when you do consider its massive state-of-the-art hardware. Don't be blind...
I think you are blinded by your interest in chess. Deepmind/Google has in my opinion no interest at all in chess. They used SF, and similar for Shogi and Go, to prove that their methods work.

They used an official download for SF and no development version, make sense to me.

Trained A0 until it was stronger and then showed the world what they can do. Don't you see the whole point? In a few hours A0 was superior to an engine developed for many years by humans. Against a stronger chess engine they had just enhanced the training of A0.

And 1 minute per move, so what? Of course SF would be stronger with another time setting but so would A0, they took the decision and played games with that settings, same for both engines right? Why is it unfair with the same setting for both engines?

Opening books? A0 did beat SF when using an opening book, not a very advanced opening book but similar to what we saw in TCEC. 50 games as white and 50 as black, what is unfair with that?

Last point the hardware difference. The complaints here are nonsense, they have showed that their combination of hardware and neural networks are better than traditional software development on traditional hardware. (Before you come shouting about other software I should add for some types of software). There's no interest in being "fair", they are interested in the commercial potential with their combination of hardware and NN and chess engines are totally uninteresting in this respect.

/John
They have showed nothing.
They made laughing stocks of themselves, comparing watermelons to blackberries.
What do you understand? The chess expert with no chess merits giving chess engine advice but don't have a clue about software development. You are the laughing stock my dear Tsvetkov.

/John

User avatar
Ovyron
Posts: 4392
Joined: Tue Jul 03, 2007 2:30 am

Re: AlphaWhat?

Post by Ovyron » Fri Dec 29, 2017 1:03 am

Lyudmil Tsvetkov wrote:25 days after the glorious event, so precisely 4*150 hours later, when Alpha should obviously be over 15-20 000 elos, STILL not a word from Google about this historic achievement...
Can you explain how 15000 elo advantage works? I don't think math allows for such an advantage unless you make intrincate things.

For instance, I don't think you can produce a virtual PGN that shows a 15000 elo advantage.

User avatar
Rebel
Posts: 5480
Joined: Thu Aug 18, 2011 10:04 am

Re: AlphaWhat?

Post by Rebel » Fri Dec 29, 2017 10:26 am

Lyudmil Tsvetkov wrote:I suggest a return match with SF being the home team.

Conditions:
- Alpha plays on 1 GPU
- a version of Alpha with no opening knowledge takes part
- TC using increment
- adapted SF plays on 2048 cores, but maybe 8192 is better
- huge HT for SF
- SF has the right of repeating the same winning opening line all over again

After the 100 game match, played at the SF premises(wherever that is, maybe somewhere on the Framework), SF team will declare the result and officially publish 10 games.

What do you think the result will be?
Nice sarcasm / cynicism but the fact the match was unfair isn't really the point. If AZ had achieved a similar result against an (ahem) 3000 elo engine it would be still the same mind blowing breakthrough.

What has been claimed by Deepmind is monstrous, they come along and wipe out 50 years of CC development by thousands of people in 4 hours of computer time without (as they claim) any domain knowledge, not even piece values.... that's beyond (my) understanding and (my) imagination and if true CC programming has changed forever, it will be NN programming from now on for those who seriously want to compete.

What's also is important is the source of the info. If it came from someone unknown it likely would have been received with laughter and jeer. But it comes from a company with a reputation of innovation and when they pick a project it's always a big challenge, as this one. It's the reputation that makes people think they are speaking the truth even if the paper needs a lot to be desired.

Rodolfo Leoni
Posts: 545
Joined: Tue Jun 06, 2017 2:49 pm
Location: Italy

Re: AlphaWhat?

Post by Rodolfo Leoni » Fri Dec 29, 2017 2:19 pm

Rebel wrote:.........................................................
What has been claimed by Deepmind is monstrous, they come along and wipe out 50 years of CC development by thousands of people in 4 hours of computer time without (as they claim) any domain knowledge, not even piece values.... that's beyond (my) understanding and (my) imagination and if true CC programming has changed forever, it will be NN programming from now on for those who seriously want to compete.

What's also is important is the source of the info. If it came from someone unknown it likely would have been received with laughter and jeer. But it comes from a company with a reputation of innovation and when they pick a project it's always a big challenge, as this one. It's the reputation that makes people think they are speaking the truth even if the paper needs a lot to be desired.
Just a conjecture from a prophane: while building its Monte Carlo statistics and propagating all those millions game scores (and maybe other parameters) back to start positions, engine could have a code section for building its own evaluation parameters too. Some kind of TD-Lambda, but from all starting parameters set to 0.

The question is: what's Deep Mind's target? It'd be a nonsense to advertise all these "events" with Shogi, Go, and Chess without having plans for delivering a product or a service. And I can't think at such a powerful hardware only for board games so I guess they plan to offer a package which could include a new OS, lots of utilities in many fields, and maybe tools for programmers. Will it be an elite product? I think it'll start as an expensive, but affordable product, as it was for Tesla cars. Then, after some years, it could become cheap enough for most people.

Meanwhile, Xeon and Ryzen CPU prices could drop...
F.S.I. Chess Teacher

Michael Sherwin
Posts: 3196
Joined: Fri May 26, 2006 1:00 am
Location: WY, USA
Full name: Michael Sherwin

Re: AlphaWhat?

Post by Michael Sherwin » Fri Dec 29, 2017 2:42 pm

Rodolfo Leoni wrote:
Rebel wrote:.........................................................
What has been claimed by Deepmind is monstrous, they come along and wipe out 50 years of CC development by thousands of people in 4 hours of computer time without (as they claim) any domain knowledge, not even piece values.... that's beyond (my) understanding and (my) imagination and if true CC programming has changed forever, it will be NN programming from now on for those who seriously want to compete.

What's also is important is the source of the info. If it came from someone unknown it likely would have been received with laughter and jeer. But it comes from a company with a reputation of innovation and when they pick a project it's always a big challenge, as this one. It's the reputation that makes people think they are speaking the truth even if the paper needs a lot to be desired.
Just a conjecture from a prophane: while building its Monte Carlo statistics and propagating all those millions game scores (and maybe other parameters) back to start positions, engine could have a code section for building its own evaluation parameters too. Some kind of TD-Lambda, but from all starting parameters set to 0.

The question is: what's Deep Mind's target? It'd be a nonsense to advertise all these "events" with Shogi, Go, and Chess without having plans for delivering a product or a service. And I can't think at such a powerful hardware only for board games so I guess they plan to offer a package which could include a new OS, lots of utilities in many fields, and maybe tools for programmers. Will it be an elite product? I think it'll start as an expensive, but affordable product, as it was for Tesla cars. Then, after some years, it could become cheap enough for most people.

Meanwhile, Xeon and Ryzen CPU prices could drop...
Hi Rodolfo, From someone that is barely alive but I can be rebuilt. IMHO, Google's plans for Deepmind go far beyond board games. They started with board games to nail down the principles of RL, MCTS and the deepmind NN and to show proof of concept before they spend billions more turning it into something that even Skynet would fear. In the Bible the Image of the Beast is an artificial construct that can learn and talk. Deepmind may be the predecessor to the Image of the Beast. Anyway Google is not interested in the paltry sums it can make by selling games, IMHO.

Look at my prediction here and tell me if I nailed it or not. I predicted A0!

http://www.talkchess.com/forum/viewtopi ... at&start=0
If you are on a sidewalk and the covid goes beep beep
Just step aside or you might have a bit of heat
Covid covid runs through the town all day
Can the people ever change their ways
Sherwin the covid's after you
Sherwin if it catches you you're through

User avatar
Rebel
Posts: 5480
Joined: Thu Aug 18, 2011 10:04 am

Re: AlphaWhat?

Post by Rebel » Fri Dec 29, 2017 3:12 pm

Michael Sherwin wrote:
Rodolfo Leoni wrote:
Rebel wrote:.........................................................
What has been claimed by Deepmind is monstrous, they come along and wipe out 50 years of CC development by thousands of people in 4 hours of computer time without (as they claim) any domain knowledge, not even piece values.... that's beyond (my) understanding and (my) imagination and if true CC programming has changed forever, it will be NN programming from now on for those who seriously want to compete.

What's also is important is the source of the info. If it came from someone unknown it likely would have been received with laughter and jeer. But it comes from a company with a reputation of innovation and when they pick a project it's always a big challenge, as this one. It's the reputation that makes people think they are speaking the truth even if the paper needs a lot to be desired.
Just a conjecture from a prophane: while building its Monte Carlo statistics and propagating all those millions game scores (and maybe other parameters) back to start positions, engine could have a code section for building its own evaluation parameters too. Some kind of TD-Lambda, but from all starting parameters set to 0.

The question is: what's Deep Mind's target? It'd be a nonsense to advertise all these "events" with Shogi, Go, and Chess without having plans for delivering a product or a service. And I can't think at such a powerful hardware only for board games so I guess they plan to offer a package which could include a new OS, lots of utilities in many fields, and maybe tools for programmers. Will it be an elite product? I think it'll start as an expensive, but affordable product, as it was for Tesla cars. Then, after some years, it could become cheap enough for most people.

Meanwhile, Xeon and Ryzen CPU prices could drop...
Hi Rodolfo, From someone that is barely alive but I can be rebuilt. IMHO, Google's plans for Deepmind go far beyond board games. They started with board games to nail down the principles of RL, MCTS and the deepmind NN and to show proof of concept before they spend billions more turning it into something that even Skynet would fear. In the Bible the Image of the Beast is an artificial construct that can learn and talk. Deepmind may be the predecessor to the Image of the Beast. Anyway Google is not interested in the paltry sums it can make by selling games, IMHO.
Aha, AZ = 666, how clever :)
Look at my prediction here and tell me if I nailed it or not. I predicted A0!
HAL on a cell phone would crush AZ in a 100 game match with 28 wins and no single loss.

brianr
Posts: 442
Joined: Thu Mar 09, 2006 2:01 pm

Re: AlphaWhat?

Post by brianr » Fri Dec 29, 2017 3:28 pm

Rebel wrote: What has been claimed by Deepmind is monstrous, they come along and wipe out 50 years of CC development by thousands of people in 4 hours of computer time without (as they claim) any domain knowledge, not even piece values.... that's beyond (my) understanding and (my) imagination and if true CC programming has changed forever, it will be NN programming from now on for those who seriously want to compete.
It may seem like magic (HT AC Clarke), and it almost is, but not quite. I have been looking at Giraffe since it came out and have some sense of the concepts, but not the implementation details (and even less as far as the math). Part of the difficulty is the rapid "churn" in machine learning moving from older toolsets and h/w, then newer frameworks that became practical on faster h/w like Theano, TensorFlow, (I prefer Keras on top), with new versions that are NOT backward compatible, plus spotty but getting better Windows support, and not least the Python 2 and 3 debacle (IHMO). In fact, I think the pace is accelerating. I have finally accepted just using Ubuntu and virtual environments to handle the numerous version requirements incompatibilities, and it does work (once you find things like pip2.7, lost a day on that one).

Nonetheless, I have done some "hello world" NNs (starting with the old FANN tools for just piece values). A simple game like tic-tac-toe or even connect4 is quite tractable with desktop h/w (a good gpu helps). There a several GitHub examples.

So, take Giraffe, generalize it, and add roughly 1,700 years of compute power and A0 seems at least plausible. Yes, CC has changed markedly, along with all games, and machine learning in general. Then again, I am old enough that I still consider my smart phone a minor miracle (after nearly 50 years in IT).

Tobber
Posts: 379
Joined: Fri Sep 28, 2012 3:53 pm
Location: Sweden

Re: AlphaWhat?

Post by Tobber » Fri Dec 29, 2017 3:42 pm

Now this is not Deep learning but it is NN if you like to play with some simple simulations.

http://www.simbrain.net/


/John

Michael Sherwin
Posts: 3196
Joined: Fri May 26, 2006 1:00 am
Location: WY, USA
Full name: Michael Sherwin

Re: AlphaWhat?

Post by Michael Sherwin » Fri Dec 29, 2017 4:30 pm

Rebel wrote:
Michael Sherwin wrote:
Rodolfo Leoni wrote:
Rebel wrote:.........................................................
What has been claimed by Deepmind is monstrous, they come along and wipe out 50 years of CC development by thousands of people in 4 hours of computer time without (as they claim) any domain knowledge, not even piece values.... that's beyond (my) understanding and (my) imagination and if true CC programming has changed forever, it will be NN programming from now on for those who seriously want to compete.

What's also is important is the source of the info. If it came from someone unknown it likely would have been received with laughter and jeer. But it comes from a company with a reputation of innovation and when they pick a project it's always a big challenge, as this one. It's the reputation that makes people think they are speaking the truth even if the paper needs a lot to be desired.
Just a conjecture from a prophane: while building its Monte Carlo statistics and propagating all those millions game scores (and maybe other parameters) back to start positions, engine could have a code section for building its own evaluation parameters too. Some kind of TD-Lambda, but from all starting parameters set to 0.

The question is: what's Deep Mind's target? It'd be a nonsense to advertise all these "events" with Shogi, Go, and Chess without having plans for delivering a product or a service. And I can't think at such a powerful hardware only for board games so I guess they plan to offer a package which could include a new OS, lots of utilities in many fields, and maybe tools for programmers. Will it be an elite product? I think it'll start as an expensive, but affordable product, as it was for Tesla cars. Then, after some years, it could become cheap enough for most people.

Meanwhile, Xeon and Ryzen CPU prices could drop...
Hi Rodolfo, From someone that is barely alive but I can be rebuilt. IMHO, Google's plans for Deepmind go far beyond board games. They started with board games to nail down the principles of RL, MCTS and the deepmind NN and to show proof of concept before they spend billions more turning it into something that even Skynet would fear. In the Bible the Image of the Beast is an artificial construct that can learn and talk. Deepmind may be the predecessor to the Image of the Beast. Anyway Google is not interested in the paltry sums it can make by selling games, IMHO.
Aha, AZ = 666, how clever :)
Look at my prediction here and tell me if I nailed it or not. I predicted A0!
HAL on a cell phone would crush AZ in a 100 game match with 28 wins and no single loss.
Ed, look at this.

Code: Select all

a 57 the year that I was born
b 60
c 63
d 66 
e 69
f 72
g 75
h 78
i 81
j 84
k 87
l 90
m 93
n 96
o 99
p 102
q 105
r 108
s 111
t 114
u 117
v 120
w 123

S 111 + h 78 + e 69 + r 108 + w 123 + i 81 + n 96 = 666
:twisted:
If you are on a sidewalk and the covid goes beep beep
Just step aside or you might have a bit of heat
Covid covid runs through the town all day
Can the people ever change their ways
Sherwin the covid's after you
Sherwin if it catches you you're through

Rodolfo Leoni
Posts: 545
Joined: Tue Jun 06, 2017 2:49 pm
Location: Italy

Re: AlphaWhat?

Post by Rodolfo Leoni » Fri Dec 29, 2017 5:32 pm

Michael Sherwin wrote:
Rodolfo Leoni wrote:
Rebel wrote:.........................................................
What has been claimed by Deepmind is monstrous, they come along and wipe out 50 years of CC development by thousands of people in 4 hours of computer time without (as they claim) any domain knowledge, not even piece values.... that's beyond (my) understanding and (my) imagination and if true CC programming has changed forever, it will be NN programming from now on for those who seriously want to compete.

What's also is important is the source of the info. If it came from someone unknown it likely would have been received with laughter and jeer. But it comes from a company with a reputation of innovation and when they pick a project it's always a big challenge, as this one. It's the reputation that makes people think they are speaking the truth even if the paper needs a lot to be desired.
Just a conjecture from a prophane: while building its Monte Carlo statistics and propagating all those millions game scores (and maybe other parameters) back to start positions, engine could have a code section for building its own evaluation parameters too. Some kind of TD-Lambda, but from all starting parameters set to 0.

The question is: what's Deep Mind's target? It'd be a nonsense to advertise all these "events" with Shogi, Go, and Chess without having plans for delivering a product or a service. And I can't think at such a powerful hardware only for board games so I guess they plan to offer a package which could include a new OS, lots of utilities in many fields, and maybe tools for programmers. Will it be an elite product? I think it'll start as an expensive, but affordable product, as it was for Tesla cars. Then, after some years, it could become cheap enough for most people.

Meanwhile, Xeon and Ryzen CPU prices could drop...
Hi Rodolfo, From someone that is barely alive but I can be rebuilt. IMHO, Google's plans for Deepmind go far beyond board games. They started with board games to nail down the principles of RL, MCTS and the deepmind NN and to show proof of concept before they spend billions more turning it into something that even Skynet would fear. In the Bible the Image of the Beast is an artificial construct that can learn and talk. Deepmind may be the predecessor to the Image of the Beast. Anyway Google is not interested in the paltry sums it can make by selling games, IMHO.

Look at my prediction here and tell me if I nailed it or not. I predicted A0!

http://www.talkchess.com/forum/viewtopi ... at&start=0
Well, I haven't much knowledge about prophecies. I stopped thinking at religion when I read about Elohim (ancient Hebrew biblical word for Gods, plural) created man in his likeness and with the use of their image (original biblical word "Tselem", meaning something that has been separated by. With modern word, DNA). Maybe Uri could confirm it or he could say these are idiocies. I'll accept any.

About Deep Mind, progress is progress; one never knows where it brings us until it does.
F.S.I. Chess Teacher

Post Reply