Lc0 ... the GOOD and the BAD

[duncan]

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 .

It used to be 200 elo gain for each ply. Do you know what the elo gain per ply is now under normal tournament conditions.?

[jp]

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!

Do you mean you cannot detect which is AB & which NN?

[jp]

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.

No human GM or IM is as bad as Lc0 at endgames.

[dkappe]

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.

No human GM or IM is as bad as Lc0 at endgames.

That’s a common misconception. Leela, in it’s various incarnations, is weak at some endgames and brutally strong at others. Why that is so and what to do about it is a point of some discussion.

[nabildanial]

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.

No human GM or IM is as bad as Lc0 at endgames.

Leela will slaughter any human in any 9-men endgames or lower. Leela might not go for the shortest win and can't close out games against SF, but that doesn't mean she's weak at endgames.

[jp]

That’s a common misconception. Leela, in it’s various incarnations, is weak at some endgames and brutally strong at others. Why that is so and what to do about it is a point of some discussion.

Are there examples on this forum of "brutally strong" endgames? (I assume this is all without TBs.)
I've seen many examples of Leela playing ridiculous moves, though of course the net ID is changing constantly.

[M ANSARI]

Lc0 plays some endgames very very poorly ... but it also plays some endgames incredibly well. I still am not sure why that is but hopefully it will get sorted out. I think it gets a bad rep as being very poor in endgames because it tends to not use the quickest win and will endlessly torture the other side by playing sub-optimal moves. I think this has to do with how the search works and how it chooses moves. If you look at MPV of Lc0 it doesn't change its mind often if it finds a good move and will tend to not even bother to look for other lines.

[dkappe]

That’s a common misconception. Leela, in it’s various incarnations, is weak at some endgames and brutally strong at others. Why that is so and what to do about it is a point of some discussion.

Are there examples on this forum of "brutally strong" endgames? (I assume this is all without TBs.)
I've seen many examples of Leela playing ridiculous moves, though of course the net ID is changing constantly.

Already ancient history, but the old 192x15 main line nets had a good grasp of certain technical endgame (without egtb).

https://lczero.libertymedia.io//2018/07/03/bill1/

[jp]

I think it gets a bad rep as being very poor in endgames because it tends to not use the quickest win and will endlessly torture the other side by playing sub-optimal moves.

Sub-optimal moves would be okay, but going around in circles, losing drawn endgames, and drawing won endgames are not so good.

Already ancient history, but the old 192x15 main line nets had a good grasp of certain technical endgame (without egtb).

https://lczero.libertymedia.io//2018/07/03/bill1/

Thanks. I'll take a look.

[M ANSARI]

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?

Oh, I would have no clue about how to do that. But I think that there is very big potential with the new hardware that is coming for AI. For a long time GPU's were considered very poor for chess as they do not do integer based calculation. However there is a huge change in how GPU's are made and a dramatic push to produce new very high performance chips for AI. The thing that makes things different is that the new chips can do "interference" or can use a knowledge base and make local deductions based on the knowledge base on the fly. This, along with the fact that the new GPU's will be able to do 100's of trillion operations per second, makes me think that this would be perfect for tactics solving. It would be nice to see an effort to make an engine using this new hardware to solve tactics ... not really play strong chess ... just solve tactics. Tactics follow some pretty basic rules ... pinned piece, back rank mate, overloaded piece, x ray, passed queening pawn etc... This could be the knowledge base and then you just try every single check first, then every single capture possible and so on. With 100's of Trillions of Operations per second, it would seem that quite a few variations can be tried out and if there is a tactical shot it will be found very quickly. Most tactics become obviously good or bad after maybe 3 or 4 moves at most ... and they follow forced lines. I don't know why, but it just seems that these new types of chips will be very good at solving these puzzles. If you listen to a GM analyze a game after it has been played, on many occasions they can feel that a certain move is tactically correct as they see or feel the "pattern". Even a weak chess player can feel there is a tactical shot in the position, but he just cannot calculate through it. I think these patterns can be "learned" and identified and then quickly probed. The first step would be to create a tactical engine that is a puzzle solver and just have it try out various tactical suites until it does very well on them ... then use that module to independently probe positions as the main engine is playing and flag a fail high move that wants to be played, or flag a pass high move that is not being considered. The second step would be to actually do this in the search of the main engine. If step 1 is done, at least this would prevent the many 1 move blunders that Lc0 is missing. I feel this must improve Lc0 by at least 50 or maybe even 100 ELO.