TalkChess.com

On cleaned WAC tactical testsuite lc0_v21_rc1 scores pretty much the same as lc0_v20.2 with ID40949. Both to 2s/position and 20s/position.
Will check games performance.

Ah, ok ... I was hoping the tactical issues, especially the very shallow tactical misses, were fixed. Maybe that is still not resolved then! I am using the same network as is playing in TCEC (32930 I think). What is strange is that on my system (RTX 2080Ti) it seems that it had completely different opinion on yesterdays critical move in the game where Lc0 lost a big advantage. No way could I get it to go Re1 and Qg6 was preferred with a big advantage, no matter how much time I let it think on the position ... yet Re1 was played in TCEC and that seemed to give several free tempi for SF to save itself. Maybe in TCEC heating issues are causing the card to throttle down??

M ANSARI wrote: ↑Mon Feb 18, 2019 9:28 am Ah, ok ... I was hoping the tactical issues, especially the very shallow tactical misses, were fixed. Maybe that is still not resolved then! I am using the same network as is playing in TCEC (32930 I think). What is strange is that on my system (RTX 2080Ti) it seems that it had completely different opinion on yesterdays critical move in the game where Lc0 lost a big advantage. No way could I get it to go Re1 and Qg6 was preferred with a big advantage, no matter how much time I let it think on the position ... yet Re1 was played in TCEC and that seemed to give several free tempi for SF to save itself. Maybe in TCEC heating issues are causing the card to throttle down??

Yes, I suspected that too. They seem to have diminished the temperatures (earlier they did have some worrying 82-83C) to 73C and said all runs normal, but the speeds they achieve are not much higher than RTX 2080Ti alone. Also, they said that for 2 GPUs (2080Ti + 2080) they use 3 threads, which I think is not enough.

Lc0 and A0 has shown, that tactics are not so important at all. They have very small impact to rating obviously. Only important in tactical test suites .

Jouni wrote: ↑Mon Feb 18, 2019 12:51 pm Lc0 and A0 has shown, that tactics are not so important at all. They have very small impact to rating obviously. Only important in tactical test suites .

Depends, for example on chosen openings. If books are used, one might come with an anti-Leela book. Yes, in general game-play, Leela would anyway be strong on adequate and fair hardware. But such activities as puzzle-solving or finding shortest mates will still be reserved to AB engines.

Jouni wrote: ↑Mon Feb 18, 2019 12:51 pm Lc0 and A0 has shown, that tactics are not so important at all. They have very small impact to rating obviously. Only important in tactical test suites .

Actually it is pretty interesting how accurately NN mimick human play. The top GM's are generally incredibly good at positional play and understanding long term advantages to a position ... but they are extremely weak (compared to engines that is) tactically. Lc0 seems to be that way as well. I still don't understand why everyone assumes that Lc0 will always be weak tactically. Personally I think Monte Carlo search should actually be much stronger in tactical suites than AB engines as it can cover many more positions quicker, but the software for that has yet to be written. All this new hardware might actually need a total re-think of how best to utilize this new hardware. I think in the next few years there will be some major breakthoughs and we will look back at the initial games of Lc0 as we did with the first games of computers in chess. The thing is that even with the obvious infancy of Lc0 and its many obvious flaws ... it can still play some incredible chess!

Jouni wrote: ↑Mon Feb 18, 2019 12:51 pm Lc0 and A0 has shown, that tactics are not so important at all. They have very small impact to rating obviously. Only important in tactical test suites .

Teichmann said that chess was 99% tactics, computers have proven that wrong. Pick a new number.

AdminX wrote: ↑Mon Feb 18, 2019 3:13 pm

Jouni wrote: ↑Mon Feb 18, 2019 12:51 pm Lc0 and A0 has shown, that tactics are not so important at all. They have very small impact to rating obviously. Only important in tactical test suites .

Teichmann said that chess was 99% tactics, computers have proven that wrong. Pick a new number.

True.

Actually, Stockfish's tactical power has been used to prepare openings lines and general analysis. The arrival of AZ and Lc0 has opened our eyes, questioning the infallibility of the tactical approach.

If You now look at TCEC 14 games without player names, can You detect who is who? I think it's not that easy at all!

M ANSARI wrote: ↑Mon Feb 18, 2019 3:09 pm I think Monte Carlo search should actually be much stronger in tactical suites than AB engines as it can cover many more positions quicker...

How do you imagine that might work? What algorithm do you have in mind?