ALPHA ZERO or CHESS GOD
Sort:
Hi guys, sorry it’s the wrong thread but where do I find players to play I can’t find a group on the app? I am playing a couple of mates I personally know but it’s not enough.
Dec 10, 2017
0
#24
Summarizing a few points made above. I'm not such a great chess player but I know software.
- Comments like "Alpha Zero got more hardware" or "Alpha Zero wasn't working on commodity hardware" are moot. Unless you've been living under a rock, the cloud *is* commodity hardware. Ask anyone who spent $49 for an Echo Dot and can now talk to their computer. Give Stockfish the same compute hardware (albeit after having their developers figure out how to make it effectively scale on more hardware), and my money is still on Alpha Zero.
- Turn based strategy games are prime fodder for superior performance by neural nets and ML. ML wins at recognition especially when you can give it a large training set and rules that indicate good/bad. You want to talk to Alexa, or have Google automatically know which of your 10,000 photos is one of your kid? Same tech.
- I agree wholeheartedly that chess as a human social interaction will always and live on and so it should!
So, maybe it's time for AlphaZero to play - either both programs use cloud-based computing with equal bandwidth, or both programs use the same fixed hardware.
In the last chess engine tournament, Stockfish came to play. Alpha-Zero did not.
The world awaits.😐
Time to be fair play. Have a look at the games. Stockfish has been totally crushed. Overwelmed.
Pushed back with undeveloped pieces, pinned on 8th rank, gasping for air. Would not even score once with white. On one occasion, AlphaZero plays a one knight+one pawn sacrifice. Not for mate. Just translated into a deadly however final bind. Stockfish could have played with an opening book. And the defeat would have been all the more evident. AlphaZero evaluation is just crushing.
I don't look at games with asymmetrical support.😐
Dec 11, 2017
0
#28
vickalan wrote:
So, maybe it's time for AlphaZero to play - either both programs use cloud-based computing with equal bandwidth, or both programs use the same fixed hardware.
In the last chess engine tournament, Stockfish came to play. Alpha-Zero did not.
The world awaits.😐
They can't run on the same hardware. AZ, being written for a neural network runs on TPU hardware, at least partially, which isn't something that Stockfish is programmed to run on, and would image the TPU hardware can't run regular workloads.
Other than limiting the hash and not allowing the engine its opening books, the specifications don't look like Stockfish was starving for resources. It was given 64 threads and was able to calculate 70 million positions per second (from the paper). AZ only looked at 80 thousand positions per second. AZ looks at positions in a fundamentally different way.
Martin_Stahl wrote:
...They can't run on the same hardware...
Oops, a minor oversight? I guess tournaments (like this) will go on without AZ?
Dec 11, 2017
0
#30
I'm sure they could interface with some system but I don't know what the requirements are for those types of competitions. They created something that uses specialized hardware to make it work; not an oversight but a way to get their AI algorithms to work they way they need them to.
Tackling chess and competing against another engine is basically a tech demo. Sure, it could have been done a little better but for an algorithm that essentially trained itself, that is a great achievement. Maybe they'll do some additional testing, giving the other engines all the bells and whistles and see what happens. Since it is a learning algorithm, it will likely only get better 
Dec 11, 2017
0
#31
Martin_Stahl wrote:
Tackling chess and competing against another engine is basically a tech demo. Sure, it could have been done a little better but for an algorithm that essentially trained itself, that is a great achievement. Maybe they'll do some additional testing, giving the other engines all the bells and whistles and see what happens. Since it is a learning algorithm, it will likely only get better
I'm waiting for the day when alpha zero will solve chess and unleash pure drawing lines before us,,
@Martin_Stahl, I agree. In the recent tournament here (chess.com) each engine was allowed its own "virtualized instance of a hyperthreaded Intel Xeon E5-2666 v3 2.90 GHz (two processors each with 18 cores) with 60.0 GB RAM"
It is interesting that Alpha Zero learned to play chess. I would like to see its "learned" knowledge configured so that it can play in fair games. Not doing that would be like sending a team to the Olympics and not competing. Other teams will stand on the podium while AZ is on the sidelines.
Sep 12, 2018
0
#33
vickalan wrote:
That's one way but not ideal. Another way is to submit their engine in a tournament with rules of fair-play, such as chess-com-computer-championship.
Maybe next year they will submit an engine. Would be fun to watch.
Here we are as expected, the Monster Lc0 has nearly approached to puberty and proving why he is unmatchable in the world of Computer Chess.
Funny!
Nov 12, 2022
0
#35
Alpha Zero was on a computer that runs a tensor processor. It was designed to virtualize the supercomputer that the program was trained on. Unlike a conventional computer tensor processors can actively be thinking of thousands of potential moves at the same time. Each tensor processor 4 on the super computer running alpha zero has its own dedicated ram giving it a total of 800 gb of ram....lol...standard computers have about 4 or 8 gb ram...The computer stockfish was placed on in the matches was below recommended specks to run stockfish so no the match was unfair, only designed to look fair by stating only speed per minute for each computer.
I have not seen a definitive description of how Stockfish was or was not limited. If it was playing without an opening book, then the demo seems pointless, and this would be a crucial thing to know. If its clock were set to fixed time per move, that makes some sense for a demonstration, but it is still important to know about if we're going to assess what happened.
Stockfish was limited because it was assigned arbitrary (and comparatively weak) hardware.
AlphaZero was assigned its prefered and much stronger hardware, processing instructions at 100s of TFLOPs.
It's like a baseball game where one side uses baseball bats, and the other side must hit the ball with a skinny twig.
Details of the match are in (this) paper.😐