The best humans play at an Elo rating of 2800. “Stockfish,” the most powerful chess engine, has an estimated rating of more than 3500.
This may be the most terrifying statement I've read all year. From parity to ovewhelming dominance in two decades. And it may just be approaching the base of the S curve. I for one welcome ...
There used to be this fantasy that the best player would be a human augmented by a computer, optimally blending their respective gifts into the best moves. Ha. Nowadays it would be like a baby in a crib helping a heavyweight champ.
Checkout Alpha Zero chess which has no other information other than the chess rules. No opening books or repetoires. It sometimes beats Stockfish like a python strangling and squeeing a prey. It makes moves that no one understands.
> 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. Stockfish needed 10-to-1 time odds to match AlphaZero.
It played Stockfish 8, which was already a year old when they played, and the match was under arguably unbalanced circumstances.
The current version of Stockfish is Stockfish 15, and although it hasn't played AlphaZero (because AlphaZero doesn't compete), it has played clones of AlphaZero such as Leela Chess Zero in TCEC which it won against consistently for the past 5 seasons.
AlphaZero beating Stockfish was more a propaganda attempt from Google to appear as a world leader in AI then an actually attempt to find the superior engine.
Stockfish is definitely strong but the way LCZ or AZ beats its opponents is totally different and foreign to humans. Stockfish has a database and a whole bunch of openings prebaked into it.
Leela Chess Zero is definitely not as strong as Stockfish 14 but still ranked #4. That's not nothingburger.
> AlphaZero beating Stockfish was more a propaganda attempt from Google to appear as a world leader in AI then an actually attempt to find the superior engine.
Agree. I feel awful that Deepmind honchos take a round table meeting once in a while, where they pick up a board or online game and totally demolish the top player just to prove how good ML algorithms are.
There used to be this fantasy that the best player would be a human augmented by a computer, optimally blending their respective gifts into the best moves. Ha. Nowadays it would be like a baby in a crib helping a heavyweight champ.