valjr92
#1
I played some of the games on my phone and some moves Stockfish made in the games that Google revealed are not the best move at all. I mean, my phone is a Moto G6 and still Stockfish quickly shows a different move from the game Google claimed Stockfish played. Unless they played stockfish in a calculator this makes no sense. I'm not saying Alpha Zero is not better but it's almost like Google rigged the results to make their computer look better.
tpr
#2
It was an elder version 10 of Stockfish and AlphaZero ran on better, dedicated hardware.
The_Human_Paradox
#3
Stockfish 12/NNUE can easily beat AlphaZero. Remember that the Stockfish vs AlphaZero happened 3 years back when Stockfish was much weaker, and on worse hardware as well. Basically AlphaZero's victory was hyped a lot, but it's research by DeepMind did help to develop engines like Stockfish NNUE and Lc0, so it is true that that research improved the playing ability of modern engines.
petri999
#4
actually it hard to say which one run on better HW. A0 used four TPU's which was about equivalent of four high end graphics cards - just using less energty do the computations.. HW for stockfish was unknow but one coudl deduct that it was 16-32 core computer. so pretty heavy stuff as well. dunno how well SF scales with extra CPU's. ANyway there is this one sacrifice A0 made that SF would find but but took 2 hours on high end machine. So it really had some insight to game. tactically A0 is not as strong as SF but makes up with insight into position.
I think it also contributed towars human understanding of chess at top level. like harry-boy attacks aimed immobilize instead of opening a line. And really showing that quality of pieces is really more important that quantity - assumin you can play so means nothing to me.
obviously it was jusr proof of concept like exercise, Chess not being particularly important. Point was that by just adding rules sames structure network can learn to play 3 games. It did learn quickly but for learnign they did use a supercomputer level amount of TPU's. On you average laptop training would have taken thousands of years.
In go they also proved that human understanding of the game is way off. at the end A0 was at least ove 2000 elo points stronger thatn any human. Not directly measurable of course but by estimation of earlier version which still could lose to a human
chesstosterone
#5
At the time, stockfish was completely incapable of playing the way A0 did, and anyone with any sense knows that. A0 changed the chess world like Hendrix did the guitar world. It's much less interesting whether stockfish should have "won" or could "beat him now". These debates are completely inane
AlphaZero is old news. It's all about MuZero now.
dboing
#7
Muzero is for rodeo chess. it is just the true focus of google. robotics conceptual AI supremacy at the corporate level. put fear in the minds. put google on top of the AI food chain.
Chess was just a spring board. But all of the above for Alpha-zero is true, no matter the motives of Google, which i may have vilified a bit too much, but that is something I think about. As they have not seem that curious about the toy problems they honed their robotics research skills on. fine, the chess community can take over, with some tuning of mindsets....
Muzero has been aimed toward dynamical and or fluctuating environments (difference here being in information feedback loop strength between action and environmental conditions). so Rodeo chess. the board is on the bull and you are to keep playing anyway. It has not improved its performance in Chess. That might be branding credibility transfer. but I have a skeptical nature.
FPS video games, 2D tactical/strategic video games (tower games?, like COH2, names escape me)...
Meriten
#8
Stockfish 12 is better than Alpha Zero and would have a higher Elo. You can prove this very easily. Stockfish 12 wins much more games against Stockfish 8 (the opponent of Alpha Zero) then Alpha Zero had won against Stockfish 8.
cycad321
#9
I agree with the Human Paradox.
Megadoggahedited
#10
They CHEATED THEIR ASSES OFF. This including running stock fish on a single core and alpha zero on a mainframe. They also removed SF opening book.
So there’s that.
