also, you farm people who are at least 100 elo below you after you started dropping. That's called "rating manipulation" :)
Ok @fpvbmct please let me know when you reach 2000s in blitz and I will be very happy to play with you all matters aside :D
yes, no hard feelings, i'll tell you for sure :)
The match wasn't "fair" however not to the point where it was actually meaningful.
A0 used TPUs, SF used CPUs. Nothing wrong with that, each are build for their specific processor type. And the 64 cores SF got weren't few either. However SF was configured poorly, most notably a small TT size.
Actually looking at the games it might not matter much, A0 showed some really deep understanding of chess. A stronger SF might have lost fewer games but it seems doubtful that it would've even won a single game. And A0 isn't optimized for chess at all, theres a lot of potential for chess programmers to improve upon that. Checkout the chess port of leelaZero.
Furthermore DeepMind doesn't need to "prove" they are stronger by playing in TCEC or anything (which wouldn't even work because A0 doesn't run on CPUs). They don't care about chess, they just demonstrated that their general approach works for any game be it Go, Chess or Shogi. And that is true, if you look at the quality of play it's way beyond human and would be even if it had lost the match.
A0 used TPUs, SF used CPUs. Nothing wrong with that, each are build for their specific processor type. And the 64 cores SF got weren't few either. However SF was configured poorly, most notably a small TT size.
Actually looking at the games it might not matter much, A0 showed some really deep understanding of chess. A stronger SF might have lost fewer games but it seems doubtful that it would've even won a single game. And A0 isn't optimized for chess at all, theres a lot of potential for chess programmers to improve upon that. Checkout the chess port of leelaZero.
Furthermore DeepMind doesn't need to "prove" they are stronger by playing in TCEC or anything (which wouldn't even work because A0 doesn't run on CPUs). They don't care about chess, they just demonstrated that their general approach works for any game be it Go, Chess or Shogi. And that is true, if you look at the quality of play it's way beyond human and would be even if it had lost the match.
AlphaZero just works like a mad chess lover who hundreds to thousands of games against itself repeatedly for 4 hours and then trying to beat Stockfish, crushing it a few minutes later.
After all, humans know openings and endgames, so AlphaZero should be able to remember... but AlphaZero plays such crazy moves all the time! Brilliant moves and sacrifices by AZ!
After all, humans know openings and endgames, so AlphaZero should be able to remember... but AlphaZero plays such crazy moves all the time! Brilliant moves and sacrifices by AZ!
Actually exactly this question is discussed in the latest NiC magazine. The interested readers can read various points why it wasn‘t fair, was it?
Probably it wasn’t equally fair, under fairer conditions AlphaZero would have won by a smaller margin. ;)
Probably it wasn’t equally fair, under fairer conditions AlphaZero would have won by a smaller margin. ;)
@Sarg0n Yeah I just finished looking at all the released games and I have to admit that Alphazero's approach to chess is just out of this world ,I have never seen any engine play such sacrificial stuff and that too elegantly.The deep positional understanding of alphazero is what separates it from other engines I suppose :)
If you let the engine run for a while it starts agreeing with some of the game's "mistakes". Move 36, for example, it says black's Kd8 was a mistake and that Ke8 was better. However, by depth 37 it prefers Kd8, and by depth 39 it strongly prefers Kd8, considering a much better move than the move the analysis recommended. Ke8 isn't even second-best at that point.
The computer analysis Lichess provides is certainly good, but it has to be done fairly quickly. The stockfish in the game looked much more deeply than the lichess analysis.
The computer analysis Lichess provides is certainly good, but it has to be done fairly quickly. The stockfish in the game looked much more deeply than the lichess analysis.
It was all a "controlled" publicity stunt by Google. They set the match parameters knowing in advance A0 was to be the victor. What we'll never know, is how much $ was paid to SF for accepting the match on googles terms.. Probably quite a bit. Deep pockets and all.
This conspiracy theory being said, the result does firmly show the progress being made in "AI" programming.
A0 will never play another gaming match, according to developers.
This conspiracy theory being said, the result does firmly show the progress being made in "AI" programming.
A0 will never play another gaming match, according to developers.
The AI solved chess, shogi and/in one go. It took it one day to overtake humans and fellow machines.
Don‘t think there’s any need to show more. No doubt it‘s reproducible at will, probably with a closer result though.
Don‘t think there’s any need to show more. No doubt it‘s reproducible at will, probably with a closer result though.
This topic has been archived and can no longer be replied to.