It seems that the more we understand the computer, the less faith we should place on some of the older programs, like stockfish. Analysing certain games where stockfish states a position is equal, we find alphazero refuting the equality to win. Also early moves which Stockfish consider as a losing position, alphazero has made a win. I think about all the games before the appearance of alphazero, and realise where we think a silicone analyst is impervious to error, we find now that newer, albeit, systems that play more humanlike, are playing successfull lines which stockfish simply does not understand. It is interesting that the more a computer plays like a human, the better the results it obtains, because in theory this keeps the romantic era of chess still alive.
What computer is used to analyze games here? And I wonder if we can apply the same logic as we placed before, would alphazero refute the innacuracies, mistakes or blunders in certain games? Worthy of consideration since it is possible that people can overtly criticise their own games as being poor when maybe they are not as bad as they think.
yeah, typo was being fixed as you posted this.
Unfortunately, I can't delete my reply which I was hoping to be a funny.
Probably would have been if I hadn't been quick on the mark ;)
Alpha Zero would be eaten alive by more modern engines: Stockfish has gained almost 300 estimated ELO compared to the stockfish version that played against Alpha Zero. Stockfish is, by all measures, the strongest engine as of now, and that is what lichess uses for analysis.
That's contradictory to what I have read. The most latest paper on alphazero by deepmind was published in 2019, and I don't recall there being a match up since then. Here I take some information from wiki
"In the final results, Stockfish version 8 ran under the same conditions as in the TCEC superfinal: 44 CPU cores, Syzygy endgame tablebases, and a 32GB hash size. Instead of a fixed time control of one move per minute, both engines were given 3 hours plus 15 seconds per move to finish the game. In a 1000-game match, AlphaZero won with a score of 155 wins, 6 losses, and 839 draws. DeepMind also played a series of games using the TCEC opening positions; AlphaZero also won convincingly.
Comparing Monte Carlo tree search searches, AlphaZero searches just 80,000 positions per second in chess and 40,000 in shogi, compared to 70 million for Stockfish and 35 million for elmo. AlphaZero compensates for the lower number of evaluations by using its deep neural network to focus much more selectively on the most promising variation.[1]"
I suppose fighting talk is OK, but I would not place my bets on stockfish just as of yet.
155 wins to 6 losses is very convincing, I think there is a stockfish version 14 now? And of course with any upgrade made by stockfish, alphazero will continue to evolve with its own updates. I think the core idea of finding the most promising solutions over analysing millions of positions is exactly why it
A) Plays more like a human, and
B) Capable of winning more games
We as humans do not calculate millions of variants from one move, we actually seek the most promising line. And for good reasons. Is it possible to seek millions of positions to accidently refute a winning line that can be calculated in much less? I think so, and calculating so deep could be a flaw to stockfish.
It is, but modern engines are also advancing. Stockfish is also pretty powerful, and from 2019 it has changed quite a bit...
Yes no doubt, as I said there is a version 14, but this brings us back to the loophole problem of why stockfish was probably losing more games in the first place. In effect, the advances to stockfish may prevail as futile since I still hold, that alphazero calculates promising lines, where as the arguably more powerful engine capable of analyzing millions of positions did not seem to make stockfish win more games. The only reason I can as of yet fathom why this is, is because analyzing millions of positions must have been working against it rather than for it. Even with more advanced versions of stockfish, if what I say is true, the end result should be roughly the same the algorithm was failing to find the winning lines because it was refuting them tens of millions of variants down the line.
