Leela, such a great engine (and innovation)?

[mwyoung]

https://youtu.be/q-q01F9Mmb4
How does it play like that without knowledge?
Doesn't it look like the most knowledgable engine there is?

Don't be fooled....It's all "Smoke & Mirrors" Neural Net size is the key....

If the illusion is real
Let them give you a ride
If they got thunder appeal
Let them be on your side

I propose a "Real Test" #1 LC0 with Minimal NN size vs Chess engine with Minimal opening book
Or Test #2 LC0 with Super NN size (like you are currently testing and posting bogus results) vs Chess engine with Extensive opening Book...Ex cerebellum Lets check the results......shall we?

Or I could test like I am testing.

Testing each engine at their best configuration.

I don't understand your irrational logic when it comes to Lc0.

I want to know what each engine is able to bring with their best configuration.

I don't care if Lc0 is using magic beans to play as good as it plays.

What most of us want to know is even with magic beans. How strong is Lc0 playing.

All you want to do is handicap Lc0 as much as possible. In the hope it will justify your irrational bias, and statements against a chess engine.

And your only issue with my testing. Is that is shows with data how irrational you are being against Lc0.

Bias against LC0? ......Hahahahah heck I've spent months setting it up downloading it and playing test games with the thing...and watching it play and overall what have I learned from watching it play? Almost nothing!! Why because it is not a 'Chess Engine' at least as far as what we have come to know what a Chess Engine Is....My optinion is that based on how it 'plays" (I use that word lightly) It's mostly a NN database.. a piece square grid and a very very simple move generator/engine its.. not playing its moves over the board but picking moves based on games already played (move grid)
The thing doesn't think...it's not really profound it does make unusual moves on occasion..It is great in Zugzwang type positions..right now LC0 is have trouble beating Bikjump in my Chess for android apk...what doe you want me to do..say its better than Stockfish or Sugar or Houdini Or Komodo? My job is not to make LC0 look GOOD or GREAT..I'm searching for the BEST Chess engines under minimal conditions..Thats It...

And yet it kills every engine in site.

And you keep proving my right, the more you write....

[mwyoung]

I want to know what each engine is able to bring with their best configuration.

The best configuration includes a Tournament Book, say, you hire Jeroen Noomen to produce a private book for Leela that goes into positions that the engine plays best, and that will be great against whatever opponent Leela is facing, then you will see her true strength. He did it for Rebel and Rybka when they were the best, so why not Leela?

The problem with TCEC, the CCRL, and other testing groups is that they're testing with generic books, and in those, 99% of the openings played are irrelevant.

Assume for a moment that 1.d4 is actually bad and Leela struggles with it or that the Sicilian is actually bad and Leela struggles with it, yet it's being forced to play into those openings and Stockfish reigns supreme, but a Tournament book would show that Leela can kick ass with just 1.e4 and 1...e5 against 1.e4, and that all the results of the generic openings she wouldn't play are irrelevant.

This is all made up, but you should get the point that minimal books or extensive opening books are not the answer.

You don't force Magnus Carlsen to forget all the chess theory he knows at move 4 and think by himself from there. You also don't force Magnus Carlsen and his opponents to play from a generic opening position they may not have much experience in. You let them play with a "Tournament Book" that is on their head thanks to their preparation. Why aren't engines treated the same?


All the testing of Stockfish and Leela has been nothing but charades, back in my day (...2007, yeah, I'm getting old) Zappa Mexico proved to be the most superior engine, over Rybka, under these conditions, proving that all the generic testing (where Rybka was the best) was a farce.

You have more faith in Jeroen's book then I have. I would not pay anything for a his books.
And that being said. Why should I, or anyone pay for Jeroen's books. When Lc0 is already beating every engine without Jeroen's books.

And Lc0 is a free engine. And why would Lc0 programmers pay for a book.

I think you should pay for Jeroen's books. Let me know how it turns out.

[supersharp77]

https://youtu.be/q-q01F9Mmb4
How does it play like that without knowledge?
Doesn't it look like the most knowledgable engine there is?

Don't be fooled....It's all "Smoke & Mirrors" Neural Net size is the key....

If the illusion is real
Let them give you a ride
If they got thunder appeal
Let them be on your side

I propose a "Real Test" #1 LC0 with Minimal NN size vs Chess engine with Minimal opening book
Or Test #2 LC0 with Super NN size (like you are currently testing and posting bogus results) vs Chess engine with Extensive opening Book...Ex cerebellum Lets check the results......shall we?
[/quote]

Or I could test like I am testing.

Testing each engine at their best configuration.

I don't understand your irrational logic when it comes to Lc0.

I want to know what each engine is able to bring with their best configuration.

I don't care if Lc0 is using magic beans to play as good as it plays.

What most of us want to know is even with magic beans. How strong is Lc0 playing.

All you want to do is handicap Lc0 as much as possible. In the hope it will justify your irrational bias, and statements against a chess engine.

And your only issue with my testing. Is that is shows with data how irrational you are being against Lc0.
[/quote]

Bias against LC0? ......Hahahahah heck I've spent months setting it up downloading it and playing test games with it

Bottom Line: LC0 based on what I've seen and viewed so far..is Not Strong...Is Not Smart...Is Not Great...Is Not Profound
And Is Not A CHESS ENGINE...I've Played them all my Friend From Fidelity
to Mephisto To Sugar..Houdini.. Rybka and SF Learning and Komodo 12....It's not even close...its a Scandal

[Ovyron]

You have more faith in Jeroen's book then I have. I would not pay anything for a his books.

It was juts an idea.

The concept stands (an ultimate chess match with a strongest book the other side doesn't know) If someone can provide a better tournament book than Jeroen, and for free, let them do it.

Also, guess I was outdated, huh? So it's Stockfish who needs the book...

My whole point is that who is the best hasn't been proven yet, and it will never happen until there's a true Chess Engine Championship match, where both sides bring the strongest hardware that they can, for both Leela and the strongest Stockfish derivative out there, and they play with a private tournament book. Or the testing groups kill the lines that the engines lose because it makes no sense they'd have played them in the first place.

[Dann Corbit]

Proof of best never comes from a championship. It takes many thousands of games to know which engine is strongest when they are close in strength. And by the time we know the answer, there is a new version anyway. A contest like tcec gives us a champion. And contests like ccrl and cegt tell us what's stronger on hardware mere mortals can afford. Both systems have great value.

IMO YMMV

[Leo]

How much money in USDs would I have to spend to have TCEC 15 lczero play on a computer at home?

[Laskos]

How much money in USDs would I have to spend to have TCEC 15 lczero play on a computer at home?

i5 + 2080ti + 2080 + 32GB RAM + accessories, about $2500 or so. Cheaper than you get SF to play at TCEC level. And if TCEC will keep the same conditions, Leela will demolish SF in SuFi 16.

[chrisw]

How much money in USDs would I have to spend to have TCEC 15 lczero play on a computer at home?

i5 + 2080ti + 2080 + 32GB RAM + accessories, about $2500 or so. Cheaper than you get SF to play at TCEC level. And if TCEC will keep the same conditions, Leela will demolish SF in SuFi 16.

and don’t skimp on the PSU.

[Leo]

https://youtu.be/q-q01F9Mmb4
How does it play like that without knowledge?
Doesn't it look like the most knowledgable engine there is?

Don't be fooled....It's all "Smoke & Mirrors" Neural Net size is the key....

If the illusion is real
Let them give you a ride
If they got thunder appeal
Let them be on your side

I propose a "Real Test" #1 LC0 with Minimal NN size vs Chess engine with Minimal opening book
Or Test #2 LC0 with Super NN size (like you are currently testing and posting bogus results) vs Chess engine with Extensive opening Book...Ex cerebellum Lets check the results......shall we?

I wont be convinced LCZ is better until it beats Brainfish. It could but lets see.

[supersharp77]

How much money in USDs would I have to spend to have TCEC 15 lczero play on a computer at home?

i5 + 2080ti + 2080 + 32GB RAM + accessories, about $2500 or so. Cheaper than you get SF to play at TCEC level. And if TCEC will keep the same conditions, Leela will demolish SF in SuFi 16.

and don’t skimp on the PSU.

I wont be convinced LCZ is better until it beats Brainfish. It could but lets see.
[/quote]

Proof of best never comes from a championship. It takes many thousands of games to know which engine is strongest when they are close in strength. And by the time we know the answer, there is a new version anyway. A contest like tcec gives us a champion. And contests like ccrl and cegt tell us what's stronger on hardware mere mortals can afford. Both systems have great value.

IMO YMMV

Dann Corbit Gives Us "The bottom Line" On This Issue....

My Take Is Minimal Engine/Program assistance with no "Cheating or Edges' in TCEC
Or If LC0 and these NN Programs are allowed Unlimited NN sizes and GPu's then Stockfish is allowed to have some sort of opening book and Huge number of cores..."Lets Level The Playing Field and Check the Results" AR