Just think, some day we may be able to say "once, I won a game against Leela, the current best player in the world".
Leela Zero( A Neural Network engine similar to Alpha Zero)
Sort:
Next should be 2800+ Carlsen.
Leela crushed 2400 IM with score 10-1. That means Leela is +400 elo better than 2400 IM. ( Hardware they used was Nvidia 980 Ti, setting 15'+5")
Apr 14, 2018
0
#145
Nah, i dont think so. She makes 2 move tactical blunders, see that black bishop pin was huge blunder in french game
I told you several times, that play.lczero.orgis for demonstration only. She think only 200 positions in normal mode and 2000 in hard mode. In that match it was full mode, she might think about 200,000-300,000 positions per turn.
BTW, it was the first public match vs Title player. You can click the game on link #155.. (There were more than 100 spectators)
Apr 14, 2018
0
#148
drmrboss wrote:
I told you several times, that play.lczero.orgis for demonstration only. She think only 200 positions in normal mode and 2000 in hard mode. In that match it was full mode, she might think about 20,000-30,000 positions per turn.
Still blundered very badly https://lichess.org/3Bl93u7n/black#84
AIM-AceMove wrote:
drmrboss wrote:
I told you several times, that play.lczero.orgis for demonstration only. She think only 200 positions in normal mode and 2000 in hard mode. In that match it was full mode, she might think about 20,000-30,000 positions per turn.
Still blundered very badly https://lichess.org/3Bl93u7n/black#84
Leela that played vs IM has 80,000 playouts per turn in average. (data from leela team). Which is 40x more searching power than Hardmode (2000 playouts) in play.lczero.org , or 400x searching power than Normal mode(200 playouts) . So estimate, 800 elo stronger than normal mode, or 500 elo stronger than hardmode. ( 2x search power = about100 elo for leela, meanwhile 2x search power=50 elo for stockfish)
Apr 15, 2018
0
#151
drmrboss wrote:
AIM-AceMove wrote:
drmrboss wrote:
I told you several times, that play.lczero.orgis for demonstration only. She think only 200 positions in normal mode and 2000 in hard mode. In that match it was full mode, she might think about 20,000-30,000 positions per turn.
Still blundered very badly https://lichess.org/3Bl93u7n/black#84
Leela that played vs IM has 80,000 playouts per turn in average. (data from leela team). Which is 40x more searching power than Hardmode (2000 playouts) in play.lczero.org , or 400x searching power than Normal mode(200 playouts) . So estimate, 800 elo stronger than normal mode, or 500 elo stronger than hardmode. ( 2x search power = about100 elo for leela, meanwhile 2x search power=50 elo for stockfish)
Good to know. But she blundered whole piece in 3 move tactic. She is not ready for Magnus. Numbers dont mean anything if real play is not on same level.
On my card Nvidia 1060 , i guess 2200. On Nvidia 1080 ti , she would be 2400.
AIM-AceMove wrote:
drmrboss wrote:
AIM-AceMove wrote:
drmrboss wrote:
I told you several times, that play.lczero.orgis for demonstration only. She think only 200 positions in normal mode and 2000 in hard mode. In that match it was full mode, she might think about 20,000-30,000 positions per turn.
Still blundered very badly https://lichess.org/3Bl93u7n/black#84
Leela that played vs IM has 80,000 playouts per turn in average. (data from leela team). Which is 40x more searching power than Hardmode (2000 playouts) in play.lczero.org , or 400x searching power than Normal mode(200 playouts) . So estimate, 800 elo stronger than normal mode, or 500 elo stronger than hardmode. ( 2x search power = about100 elo for leela, meanwhile 2x search power=50 elo for stockfish)
Good to know. But she blundered whole piece in 3 move tactic. She is not ready for Magnus. Numbers dont mean anything if real play is not on same level.
Man, your argument seemed more and more similar to Lyudmil . (He always says, stockfish is weak and he can beat SF 8-2. Actually the version of stockfish he played was version 4 or 5 with depth under 20 ply).
Apr 15, 2018
0
#155
drmrboss wrote:
AIM-AceMove wrote:
drmrboss wrote:
AIM-AceMove wrote:
drmrboss wrote:
I told you several times, that play.lczero.orgis for demonstration only. She think only 200 positions in normal mode and 2000 in hard mode. In that match it was full mode, she might think about 20,000-30,000 positions per turn.
Still blundered very badly https://lichess.org/3Bl93u7n/black#84
Leela that played vs IM has 80,000 playouts per turn in average. (data from leela team). Which is 40x more searching power than Hardmode (2000 playouts) in play.lczero.org , or 400x searching power than Normal mode(200 playouts) . So estimate, 800 elo stronger than normal mode, or 500 elo stronger than hardmode. ( 2x search power = about100 elo for leela, meanwhile 2x search power=50 elo for stockfish)
Good to know. But she blundered whole piece in 3 move tactic. She is not ready for Magnus. Numbers dont mean anything if real play is not on same level.
Man, your argument seemed more and more similar to Lyudmil . (He always says, stockfish is weak and he can beat SF 8-2. Actually the version of stockfish he played was version 4 or 5 with depth under 20 ply).
What part of she blundered whole piece in simple tactic you dont understand?
She missed simple tactics cos you played in low ply demonstration server. with 200, 2000 nodes. Quite simple, if you do stockfish with 6 ply depth, 3500 SF will miss such simple tactics.
Leela (2400) vs scorpio (2990 ) on TCEC is live now .Please not leela is playing without GPU, she is severly handicapped.
Apr 15, 2018
0
#159
drmrboss wrote:
She missed simple tactics cos you played in low ply demonstration server. with 200, 2000 nodes. Quite simple, if you do stockfish with 6 ply depth, 3500 SF will miss such simple tactics.
The blunder was vs the IM guy.
Leela is beating IM easily. 4.5-0.5, the game is going on live. OMG!!
https://lichess.org/0AvtMYAI