On December 10 2017 08:16 mishimaBeef wrote: alphazero seems moreso a player, not an evaluation engine... you don't really know if what it's doing is better or worse until it secures the result. with a traditional engine, you can play a few moves from a position and read the evaluation (oh +1.6, this is obviously a good spot) but with alphazero, i dunno, it seems you would have to play out the game against the best known alternative.
chess enthusiasts were using stockfish to analyze the positions in the games. kinda funny eh?
also, i think the ball is now in stockfish's court to call for a rematch
Alphazero’s long term positional sacrifices and strategic play are simply freakish « take my knight, you’ll never really be able to develop yours » and « here is a juicy pawn now good luck developping that bishop - roflcopter QID sucks » are not something you see from computers playing, usually.
That stockfish could have done better if, if and if is really not relevant. The impressiveness is qualitative, not quantitative. Alpha looks like it’s thinking. And that’s a gigantic leap forward that non chess players can’t fully appreciate. In chess terms, it totally passes the Turing test for me.
To be honest "it looks like it's thinking" is a bunch of human nonsense. Your brain is a machine, too. There's no magical line you can draw where a method suddenly qualifies as 'thinking' and becomes better for it, the causative relationship goes the other way. Thinking is an abstraction and an illusion. Ultimately the only reasonable measure of 'thinking' quality is results.
An example of this is tablebases. A complete tablebase and an efficient way to look it up is just better than a neural network thinking out an endgame, no matter how good and legitimate its method of thinking is.
Don't get me wrong, I fully agree neural networks are a big leap forward, I just don't like the mysticism.
The interesting thing for me is that Alpha Zero seems to have a positional understanding of the game (i.e. pattern recognition) in that its playing for long-term compensation where the possible variations are computationally infeasible to calculate fully.
Obviously its better to use table base where every single possibility has been mapped out but you dont have a 32 piece table base to rely on when the game starts. The games where AZ won were mostly decided in the middle game when its not possible to look 100 moves ahead.
On December 24 2017 09:34 Lachrymose wrote: To be honest "it looks like it's thinking" is a bunch of human nonsense. Your brain is a machine, too. There's no magical line you can draw where a method suddenly qualifies as 'thinking' and becomes better for it, the causative relationship goes the other way. Thinking is an abstraction and an illusion. Ultimately the only reasonable measure of 'thinking' quality is results.
An example of this is tablebases. A complete tablebase and an efficient way to look it up is just better than a neural network thinking out an endgame, no matter how good and legitimate its method of thinking is.
Don't get me wrong, I fully agree neural networks are a big leap forward, I just don't like the mysticism.
so do you think alphazero has the self-conscious "illusion of thinking" like we do?
Back on AZ has anyone seen this analysis of game 10? Its incredible chess, it really is. Early piece and pawn sacs open up long diagonal lines that are used to handcuff the queens side and open up the kings side simultaneously by using odd looking queen moves (Qh1!)
Starting from round 3, Svidler will be on the chess24 stream together with Gustafsson. I don't think there's anyone who can match Svidler as a commentator, so it's what I'll be watching.
Seirawan will be commentating on the chessbrah stream on twitch with Eric Hansen and all the rest. Seirawan's great and their commentary is always very enthusiastic and entertaining.
I'm sure the official stream with Polgar's good too, but you have to pay to watch, so meh.
Not sure about streams for newer players, but I'm guessing there are plenty.
Yes at Grenke, I'm the webmaster guy, also I do a lot of stuff before and during the tournament that just keeps the tournaments running. I'll be on location during the big open.
On March 14 2018 05:05 don_kyuhote wrote: Came back from a chess tournament. Beat two 1900s uscf and finally broke into the 1800s, phew. One 100s at a time. Next goal is 1900 uscf.
Kramnik's end of match analysis has been really... interesting. He's known for being overly optimistic about his chances, but the last two games have really been something beyond the usual. Ding handled it well.
AZ is absolutely amazing,they did beat go which I thought would be at least 20+ years away and now they are onto chess.Am curious to see how far it will go and that game was awesome,thanks for linking it.