Practical implication of LCZero on future human chess play?

[ad]

Can somebody try to give an insight into the practical side of human OTB chess evolution?
Can super grand masters already use leela chess zero for their training?

[Nathan Blaxall]

This is already happening, and helping top players and also "average" players too.

Leela has discovered new opening lines (finding novelties, and refuting others) so that now anyone can use those opening lines.

Some strong players are using stockfish and Leela side by side as they prepare for their next tournament.

Also, since Leela cares about position more than counting material, it's helping people to do the same (eg more consideration given to giving up material in order to block in enemy pieces).

From a non-human angle, I am supposing that the stockfish devs now have (some) new information & ideas to tweak & test to get stockfish stronger.

[Dietrich Kappe]

Is this speculation or informed speculation? I know of no super gm’s and their seconds buying gpu’s to run leela? Can you point me to some stories claiming otherwise?

[Ron Langeveld]

I can but I don't want to mention any names for obvious reasons.

[Tryfon Gavriel]

I think we can track Leela's influence on OTB play by tracking the amount of winning or losing "thorn pawn" games as one new "feature" of modern OTB chess - right from Opening innovations, to combinations generally in the middlegame. 

It seems Leela is constantly using Thorn pawns in many winning games. If Leela is influencing OTB players, I suspect this is just one aspect. A bigger list may be:

1) More "Thorn" pawns

2) More dynamic and well timed pawn breaks

3) More dynamic and aggressive use of pawns in general (Kasparov, Topalov already had this in his style!). Kasparov also frequently uses h5-h4 type attacks as well. 

4) Attacks which seem a little unsound at first but have passed pawn potential for endgame transition

5) Gambits which aim to lock in the c8 (c1) bishop - "Alpha zero" stylistic feature too

2-4 echo Philidor's "The pawns are the soul of chess"

Cheers, K

[Nathan Blaxall]

I wouldn't know about super gm's (I was only talking about "strong" players), but I've read someone saying they were using Leela & stockfish side-by-side to help prep for their next tournament, so I'm assuming they are a strong player ... forget where I read that.

For new opening theory, I was thinking of this: https://www.youtube.com/watch?v=0Hhhpu2_Ugk

[Matt Blakely]

I am literally waiting for the day superGMs start to use Leela - the opportunity for new ideas is INSANE!

Leela has the chance to revitalize top chess in many ways, as openings and play have gone a bit stagnant, and new ideas are rare.  But leela shows us new ideas that top players can copy and play.

I follow human chess closely too and will report if I see confirmation of a top GM using Leela or similar.  

But given her youth and the desire for advantage, I do'nt think players will want to tell people this.  So just because we haven't heard it doesn't mean it isn't happening.

[Jonasobv]

 I bet many chess players already more or less use leela for their prep. Especially on top level. I mean it would be kind of stupid not to. Leela is showing very interesting opening ideas and is starting to beat sf9. Especially considering that she often gets very good positions out of the opening.

Even Jon-Ludvig Hammer stated in his youtube video that he is following leelas progress and that was a few months back if i remember correctly, and leela wasn't even close to its current strength. So why would he suddenly not care anymore. Engines play a huge role in todays otb chess and their prep. I think as a Super GM or especially as a second you are kinda forced to follow the current meta on chess-engines. You take every advantage you can get and surprising your opponent with a new line is simply gold. I think you just don't want to give away your secrets so easily, thats maybe why we don't hear about it too often. From watching modern GMs analysing a chess position they even know very well in what positions they can trust an engine and what different engines are good at.

Apart from that, if you love chess you would simply be curious about how a NN would evaluate your favourite openings and weapons.

[Craig Campbell]

I would like to know names as I find this hard to believe. Lc0 doesn't scale at all. On my 1080 gpu it has made as much progress into the position as it is going to after about 10 seconds. Its ending play is feeble by professional standards. TBs won't solve this because lc0 is looking 20 ply deep when the brute force bots are looking 50. It might come up with a new opening idea every so often but the brute force bots do it all the time. From the human perspective lc0 has a long way to go IMO.

[Dietrich Kappe]

Hammer also criticized a lack of multipv. Still not fixed

[ccamp81318]

Sorry I didn't mention what I consider Leela's strong points. She does have a different perspective than the brute force bots which is of value. Even though I think she has a long way to go to be competitive at real TCs if you are running a cuda version she costs very little to run simultaneously with your favorite bot (with an adequately cooled rig).

A question regarding Leela's search depth: It seems that the greatest gains by the brute force bots has come since they started to work on their front end pruning. Perhaps Leela could be more vigorous in this area. As I said She very rarely changes her favorite candidate after the first few seconds. Just an idea.

[Nathan Blaxall]

Let's note that (I would say) 99% of a gm's tournament prep is opening prep.  They don't play again a bot for practice - that's not how they prep.

As said, if they can find an opening novelty, it's gold.  It doesn't have to be a move that refutes the opening line either, it can just be a surprising (but sound) move that the opponent didn't prep for.  The opponent will then have to spend lots of clock time OTB to get to grips with it, and that advantage can make all the difference.

This is what GM's battle to do all the time.

Also, let's note that super GM's have a team of seconds.

They would be crazy not to use Leela as just one of their tools (and they'd have many tools).

They would go even further and make sure they were familiar with novel lines Leela might come up with against their opening prep, in case their opponent used Leela for their opening prep.

I'd be astounded if Leela isn't being used all over the place at top level chess (even as she is right now).

[Dietrich Kappe]

Good enough for me. If this is the case, I wonder if we’ll hear about some TN due to leela being uncorked in a tournament.

[frakty]

Well there is a video of top players answering various question about A0 when it became a big news

https://www.youtube.com/watch?v=M1Dva2KCmBw

at 9:00 Segey Karjakin says that he would be willing to pay 100k $ for A0 to have it "today"

And nowdays on good hardware LC0 is becoming close to A0 level