No. Not in the long run.
stocfish-12 vs alphazero
Sort:
Stockfish is an open source project headed by a small but talented team. Alphazero was a project headed by one of the largest corporations in the world with deep pockets and the ability to hire anyone they want.
That said, Alphazero is not an ongoing project and google doesn’t devote resources to its development. Chess for them was just a proof of concept and no more and they don’t really care about the game. In their eyes, they have nothing to gain.
So the question is sort of irrelevant. If google wanted to, they could easily make the best engine in the world by far, but they don’t have any desire to do so.
That said, Alphazero is not an ongoing project and google doesn’t devote resources to its development. Chess for them was just a proof of concept and no more and they don’t really care about the game. In their eyes, they have nothing to gain.
So the question is sort of irrelevant. If google wanted to, they could easily make the best engine in the world by far, but they don’t have any desire to do so.
Yeah same here @Sameer_acchab
Stockfish 11 beat Lc0 in TCEC 18. Lc0 is basicaly A0 on steroids, strength- wise, and then you have Stockfish 12 which is even stronger than Stockfish11. On top of that you have the dev. version of Stockfish which is 20-30 elo stronger than S12, and is currently leading in a match against the latest dev. version of Leela by one point in the TCEC 19 superfinal. So yeah, it's safe to say that Alpha zero would get it's ass whooped in a match against the latest version of Stockfish.
where do you guys follow the engine games updates?
stockfish 15 would kill alphazero 1,000 times in a row before a draw
AA-AAchannel wrote:
stockfish 15 would kill alphazero 1,000 times in a row before a draw
Would be interesting to see, though it's unlikely they'll revisit.
AA-AAchannel wrote:
stockfish 15 would kill alphazero 1,000 times in a row before a draw
AlphaZero's magic was that it was a self-taught, self-learning entity. It reached its pinaccle by only playing itself. It had no other source of knowledge, other than self-play.
It played against itself 44 million times.
Now ... imagine how strong it would be if you had it train against Stockfish an additional 44 million times. Suddenly, it would have this new source of knowledge to build upon: an actual opponent, with a different style of play, to learn and evolve from ...
your point ironsteam
Martin_Stahl wrote:
AA-AAchannel wrote:
stockfish 15 would kill alphazero 1,000 times in a row before a draw
Would be interesting to see, though it's unlikely they'll revisit.
Indeed but a big youtube channel could ask thousand to make one game using two tabs
It is amazing to me that even today. People believe that Alpha Zero is the best chess engine of all time, and is still the strongest chess engine today.
This shows you what a good job Deepmind did in the press releases. By cherry picking stats to promote to the public. And by only counting wins and losses of a 1000 game match and ignoring the draws.
But even under the best results against Stockfish 8 playing without a book, or proper time controls, and sub optimal hash table settings. Alpha Zero was still less then 50 Elo points better then Stockfish 8. With Stockfish 8 playing on weaker hardware.
When you look at the match played in 2018 were Stockfish 8 was able to play with a opening book against Alpha Zero. And this is important as Stockfish 8 is not a neural network chess engine. But a hand coded chess engine. So Stockfish 8 does not have the advantage of knowing how to play the openings like a neural network chess engine.
Alpha Zero results got even worse playing Stockfish 8. Here are the results and Elo of that 100 game match played with opening books, but still with Stockfish 8 playing Alpha Zero with sub optimal settings on weaker hardware.
Computer Match 2018
1 AlphaZero +31 +17/=75/-8 54.50% 54.5/100
2 Stockfish 8 -31 +8/=75/-17 45.50% 45.5/100
Now Alpha Zero in this more proper match setup with opening books playing Stockfish 8. Alpha Zero lost 8 games to Stockfish 8. Not the more widely reported and only remembered 6 losses in a 1000 games.
Now Alpha Zero beat Stockfish by only 31 Elo points.
So the conclusion that Alpha Zero is the best chess engine of all time is clearly wrong. And by the Elo statistics Alpha Zero is weaker then today's chess engines Dragon, Stockfish, Lc0, Berserk 10, Koivisto 9, Revenge 3, Rubichess, Ethereal, Slowchess Blitz, Seer, Rebel 16, Rofchade, Minic, Igel, Uralochka, Arasen, and Ginkgo Vienna.
AArjun1516 wrote:
stockfish 15 would kill alphazero 1,000 times in a row before a draw
I mean .... alphazero .... Bro was bullying stockfish then.
For anyone interested, the DeepMind team stated that, if they ran the AlphaZero algorithm on modern hardware, it would beat AlphaZero from the past by 100-0 ...
They said this process would continue indefinitely into the future, as long as hardware continues to advance - claiming that the AZ algorithm would perpetually take chess "to the highest levels of knowledge that you can achieve in a domain". (https://www.chess.com/news/view/david-silver-alphazero-reinforcement-learning)
So, looking at the strength of AZ from the past won't give us an accurate picture, as the team declared that AZ will continually play stronger and stronger with each newer iteration.
So ... if you ran AZ today, it would apparently walk all over the previous version of itself with relative ease.
AArjun1516 wrote:
stockfish 15 would kill alphazero 1,000 times in a row before a draw
If you’ve seen the games, AlphaZero DEMOLISHED Stockfish 16 in many games, one in which AlphaZero completely trapped Stockfish’s in the corner
DeathKiller1987 έγραψε:
AArjun1516 wrote:
stockfish 15 would kill alphazero 1,000 times in a row before a draw
If you’ve seen the games, AlphaZero DEMOLISHED Stockfish 16 in many games, one in which AlphaZero completely trapped Stockfish’s in the corner
Do you own a time machine?
Stockfish 16 was out at 2023-06-30, while Alphazero has not played any games since early 2017.
If not, then stop claiming FUD.
Towar60 έγραψε:
AArjun1516 wrote:
stockfish 15 would kill alphazero 1,000 times in a row before a draw
I mean .... alphazero .... Bro was bullying stockfish then.
Stockfish 8 (non NNUE) running misconfigured on average hardware. Not such a big deal, actually Stockfish 8 was running below its playing strength on purpose.
do you think that if a fight is conducted between alpha zero and stockfish-12 then stockfish-12 would win?