Amazing results of LeelaKnightOdds-Dev

[lkaufman]

I guess that there are people who are weaker against leela because they believe that she cannot blunder so do not care to calculate if some sacrifice is leading to a forced mate against leela in position it is obvious there was not a forced mate that humans can calculate.

Of course it is correct for normal AB engines but not for leela and the shortest loss of leela queen odds bot is because leela let a mate that humans could calculate.

https://lichess.org/2gk2PfYa/black

I do not blame leela for playing 9...Rg8 and not 9...Rf8 10.Nf7+ Rxf7 11.Qxf7 that does not give practical chances.

Basically (oversimplifying a bit), the Leela odds bots "assume" that they are playing an opponent who is just strong enough to score about 50% against Leela at the stated odds, and play accordingly, not worrying about things that such a player is not expected to see. If you can see things that players of the targeted level cannot see, you will usually win. Pretty much the same way a chess hustler playing for money would play!

[Father]

I guess that there are people who are weaker against leela because they believe that she cannot blunder so do not care to calculate if some sacrifice is leading to a forced mate against leela in position it is obvious there was not a forced mate that humans can calculate.

Of course it is correct for normal AB engines but not for leela and the shortest loss of leela queen odds bot is because leela let a mate that humans could calculate.

https://lichess.org/2gk2PfYa/black

I do not blame leela for playing 9...Rg8 and not 9...Rf8 10.Nf7+ Rxf7 11.Qxf7 that does not give practical chances.

Basically (oversimplifying a bit), the Leela odds bots "assume" that they are playing an opponent who is just strong enough to score about 50% against Leela at the stated odds, and play accordingly, not worrying about things that such a player is not expected to see. If you can see things that players of the targeted level cannot see, you will usually win. Pretty much the same way a chess hustler playing for money would play!

I wonder... in case I reach a performance of, for example, 40% like this: 6 draws, 4 losses, what would happen to my elo? ...also the
past, past this, the future does not yet exist, this means that a human after having a very extensive losing streak, can take the measure of the ferocious and indomitable equine, the
entity, tame it a bit, and you get closer to its performance. Fixing present results to the prediction of the past is statistical nonsense.

[Uri Blass]

I guess that there are people who are weaker against leela because they believe that she cannot blunder so do not care to calculate if some sacrifice is leading to a forced mate against leela in position it is obvious there was not a forced mate that humans can calculate.

Of course it is correct for normal AB engines but not for leela and the shortest loss of leela queen odds bot is because leela let a mate that humans could calculate.

https://lichess.org/2gk2PfYa/black

I do not blame leela for playing 9...Rg8 and not 9...Rf8 10.Nf7+ Rxf7 11.Qxf7 that does not give practical chances.

Basically (oversimplifying a bit), the Leela odds bots "assume" that they are playing an opponent who is just strong enough to score about 50% against Leela at the stated odds, and play accordingly, not worrying about things that such a player is not expected to see. If you can see things that players of the targeted level cannot see, you will usually win. Pretty much the same way a chess hustler playing for money would play!

I wonder... in case I reach a performance of, for example, 40% like this: 6 draws, 4 losses, what would happen to my elo? ...also the
past, past this, the future does not yet exist, this means that a human after having a very extensive losing streak, can take the measure of the ferocious and indomitable equine, the
entity, tame it a bit, and you get closer to its performance. Fixing present results to the prediction of the past is statistical nonsense.

6 draws and 4 losses are 3 point out of 10 so 30%.

In this case by the rating rules your 6 draws count as 3 draws so practically it is the same as getting 1.5 out of 7 that is near 21.4%

21% translate to 230 elo difference and 22% to 220 elo difference so practically you will get performance of 226 elo weaker than the bot based on the unfair calculation.

LeelaQueenOdds as black 1+0 (Reference) has a rating of 2795.2 so you can expect to improve your rating assuming it is smaller than 2569 in this case.

[Uri Blass]

I guess that there are people who are weaker against leela because they believe that she cannot blunder so do not care to calculate if some sacrifice is leading to a forced mate against leela in position it is obvious there was not a forced mate that humans can calculate.

Of course it is correct for normal AB engines but not for leela and the shortest loss of leela queen odds bot is because leela let a mate that humans could calculate.

https://lichess.org/2gk2PfYa/black

I do not blame leela for playing 9...Rg8 and not 9...Rf8 10.Nf7+ Rxf7 11.Qxf7 that does not give practical chances.

Basically (oversimplifying a bit), the Leela odds bots "assume" that they are playing an opponent who is just strong enough to score about 50% against Leela at the stated odds, and play accordingly, not worrying about things that such a player is not expected to see. If you can see things that players of the targeted level cannot see, you will usually win. Pretty much the same way a chess hustler playing for money would play!

I believe that humans can improve against the bot.
Humans who prepare seriously against the bot need only to learn from analyzing games what the bot expect you not to see and simply learn to see it fast.

I believe that being able to get more than 50% against the bot at 1+1 should be an easier task then getting the IM title.

Even if IM's today do not get 50% against the bot at 1+1 then I believe the reason is that they did not spend signficant time to learn to beat the bot and used most of their time for normal chess.

[Uri Blass]

leader board with some details about the leaders.

LeelaQueenOdds as white 1+0 (Reference) 2895.2
LeelaQueenOdds as black 1+0 (Reference) 2795.2
LeelaQueenOdds as white 1+1 (Reference) 2634.5
LeelaQueenOdds as white 3+0 (Reference) 2619.5
LeelaQueenOdds as black 1+1 (Reference) 2534.5
LeelaQueenOdds as black 3+0 (Reference) 2519.5
1 Master_Unknown 2514.6 92
bullet rating 2881 with white won 3 games with white at 3+0 and one with white at 1+1
with black lost 19-1 with 2 draws at 1+0.
result 23 wins 23 losses and 6 draws with 1+1
2 Orichess88 2481.1 42(not very active in lichess so no stable rating)
always black
won 29 games lost 10 games and drew 3 games at 2+3
lost one 1+0 game.
LeelaQueenOdds as white 3+2 (Reference) 2450.0
3 Ushkothebear 2445.6 184 blitz rating 2530

played 184 games against Leela with LeelaQueenOdds as white 5+3(lost 87 games won 87 games and 10 draws)

4 IM Omnivor 2396.8 30 blitz rating 2406 fide rating 2404 based on the page.
played 10 games with black at 5+5 and won 8.5-1.5(draw by stupid mistake of making stalemate by playing too fast when the opponent has only a king)
played 20 games with white at 5+5 and won 19.5-0.5


LeelaQueenOdds as white 5+3 (Reference) 2353.7
LeelaQueenOdds as black 3+2 (Reference) 2350.0

5 Hissha 2339.1 85
6 Catecan 2310.8 1223
7 Respectful_Dave 2299.6 74
8 itore 2297.1 43
LeelaQueenOdds as black 5+3 (Reference) 2253.7
9 Spreek 2229.8 45
LeelaQueenOdds as white 10+5 (Reference) 2220.3
10 nn1133 2212.6 6346
11 uriblass 2211.8 165
12 AxelBoldt 2209.2 222
13 Forgive1219 2202.5 189
14 SilverMitt 2200.0 25
15 iWannaSeeYouSmile 2199.8 73
16 Ajktulm 2193.6 228

[Father]

leader board with some details about the leaders.

LeelaQueenOdds as white 1+0 (Reference) 2895.2
LeelaQueenOdds as black 1+0 (Reference) 2795.2
LeelaQueenOdds as white 1+1 (Reference) 2634.5
LeelaQueenOdds as white 3+0 (Reference) 2619.5
LeelaQueenOdds as black 1+1 (Reference) 2534.5
LeelaQueenOdds as black 3+0 (Reference) 2519.5
1 Master_Unknown 2514.6 92
bullet rating 2881 with white won 3 games with white at 3+0 and one with white at 1+1
with black lost 19-1 with 2 draws at 1+0.
result 23 wins 23 losses and 6 draws with 1+1
2 Orichess88 2481.1 42(not very active in lichess so no stable rating)
always black
won 29 games lost 10 games and drew 3 games at 2+3
lost one 1+0 game.
LeelaQueenOdds as white 3+2 (Reference) 2450.0
3 Ushkothebear 2445.6 184 blitz rating 2530

played 184 games against Leela with LeelaQueenOdds as white 5+3(lost 87 games won 87 games and 10 draws)

4 IM Omnivor 2396.8 30 blitz rating 2406 fide rating 2404 based on the page.
played 10 games with black at 5+5 and won 8.5-1.5(draw by stupid mistake of making stalemate by playing too fast when the opponent has only a king)
played 20 games with white at 5+5 and won 19.5-0.5


LeelaQueenOdds as white 5+3 (Reference) 2353.7
LeelaQueenOdds as black 3+2 (Reference) 2350.0

5 Hissha 2339.1 85
6 Catecan 2310.8 1223
7 Respectful_Dave 2299.6 74
8 itore 2297.1 43
LeelaQueenOdds as black 5+3 (Reference) 2253.7
9 Spreek 2229.8 45
LeelaQueenOdds as white 10+5 (Reference) 2220.3
10 nn1133 2212.6 6346
11 uriblass 2211.8 165
12 AxelBoldt 2209.2 222
13 Forgive1219 2202.5 189
14 SilverMitt 2200.0 25
15 iWannaSeeYouSmile 2199.8 73
16 Ajktulm 2193.6 228

Good morning Mr. Larry Kaufman, Uri Blass, Marcus91, Brunetti.
I hope you are doing well and continuing very well. Life has given us a new day today. Thanks to Our common programmer, teacher and designer for life and for our existence and human algorithms, thanks to life for life.
Today I would like to ask you a special favor for me. As you well know I have lost all interest in continuing to play with the bot, due to the lack of Elo incentives. It does not mean in any way my loss of interest in facing computers. In fact, I will continue in the rings on the corner of the street in my neighborhood, beating myself with my bare hands. Since I am not a mathematician, I would like to ask you for the kind favor of calculating the ratings of the top 20 of the LeelaQueenOdds top 100 table, based on the application of the standard rules of a 0.5 draw and 1 point for a win, based on the reference and control indices.

I thank you in advance for the deference you want and wish to have for me. Again everyone, thanks for the boots. For me it is important around my training system to know my real playing strength against LeelaqueenOddds, it is essentially a matter of self-esteem and mental strength.

[Uri Blass]

leader board with some details about the leaders.

LeelaQueenOdds as white 1+0 (Reference) 2895.2
LeelaQueenOdds as black 1+0 (Reference) 2795.2
LeelaQueenOdds as white 1+1 (Reference) 2634.5
LeelaQueenOdds as white 3+0 (Reference) 2619.5
LeelaQueenOdds as black 1+1 (Reference) 2534.5
LeelaQueenOdds as black 3+0 (Reference) 2519.5
1 Master_Unknown 2514.6 92
bullet rating 2881 with white won 3 games with white at 3+0 and one with white at 1+1
with black lost 19-1 with 2 draws at 1+0.
result 23 wins 23 losses and 6 draws with 1+1
2 Orichess88 2481.1 42(not very active in lichess so no stable rating)
always black
won 29 games lost 10 games and drew 3 games at 2+3
lost one 1+0 game.
LeelaQueenOdds as white 3+2 (Reference) 2450.0
3 Ushkothebear 2445.6 184 blitz rating 2530

played 184 games against Leela with LeelaQueenOdds as white 5+3(lost 87 games won 87 games and 10 draws)

4 IM Omnivor 2396.8 30 blitz rating 2406 fide rating 2404 based on the page.
played 10 games with black at 5+5 and won 8.5-1.5(draw by stupid mistake of making stalemate by playing too fast when the opponent has only a king)
played 20 games with white at 5+5 and won 19.5-0.5


LeelaQueenOdds as white 5+3 (Reference) 2353.7
LeelaQueenOdds as black 3+2 (Reference) 2350.0

5 Hissha 2339.1 85
6 Catecan 2310.8 1223
7 Respectful_Dave 2299.6 74
8 itore 2297.1 43
LeelaQueenOdds as black 5+3 (Reference) 2253.7
9 Spreek 2229.8 45
LeelaQueenOdds as white 10+5 (Reference) 2220.3
10 nn1133 2212.6 6346
11 uriblass 2211.8 165
12 AxelBoldt 2209.2 222
13 Forgive1219 2202.5 189
14 SilverMitt 2200.0 25
15 iWannaSeeYouSmile 2199.8 73
16 Ajktulm 2193.6 228

Good morning Mr. Larry Kaufman, Uri Blass, Marcus91, Brunetti.
I hope you are doing well and continuing very well. Life has given us a new day today. Thanks to Our common programmer, teacher and designer for life and for our existence and human algorithms, thanks to life for life.
Today I would like to ask you a special favor for me. As you well know I have lost all interest in continuing to play with the bot, due to the lack of Elo incentives. It does not mean in any way my loss of interest in facing computers. In fact, I will continue in the rings on the corner of the street in my neighborhood, beating myself with my bare hands. Since I am not a mathematician, I would like to ask you for the kind favor of calculating the ratings of the top 20 of the LeelaQueenOdds top 100 table, based on the application of the standard rules of a 0.5 draw and 1 point for a win, based on the reference and control indices.

I thank you in advance for the deference you want and wish to have for me. Again everyone, thanks for the boots. For me it is important around my training system to know my real playing strength against LeelaqueenOddds, it is essentially a matter of self-esteem and mental strength.

I see that there was some mistake in my calculation so I edit it and will post again.

[Uri Blass]

To summerize the final point is that you probably lose something near 100 elo from the unfair rating list.
You also lost something near 50 elo from playing with black in big part of your games instead of playing with white that is better for rating.

It seems that with fair rating list and wise choice to play only white you could get probably rating of 2450-2500 that is not enough for first place in the list.

Earlier I made a mistake when I did not consider that you get 1/4 out of 1/2 instead of 1/4 out of 1 but it did not change a lot.

More details:
I see that from 20.11.2024 you got 123 draws with white out of 411 games and 41 draws out of 540 games with black.
some of them were 1+1 but most of them are 1+0.
It seems that some bug in the search prevent me to count the number of 1+0 games and I did not like to count manually.

LeelaQueenOdds as black 1+0 (Reference) 2795.2

123 divided by 822 possible half points is 0.1496...
Assuming you get 15% against the bot then you should be 296 elo weaker that means rating of 2499.2
with the unfair calculation you get 123/4 points out of possible 123/2+(411-123) points or 123/(246+288*4)=0.0879 and it means something like 390 elo weaker than the bot and it means performance near 2405.2.

LeelaQueenOdds as white 1+0 (Reference) 2895.2
41 draws out of 540 is 41/1080 that is 0.03796
It means performance of more than 500 elo weaker than your opponent so with fair rating list you may get something like rating of 2390 instead of something near 2300 that you get with 41/4 out of 41/2+499 that is 41/(82+1996)=0.01973...

[Father]

To summerize the final point is that you probably lose something near 100 elo from the unfair rating list.
You also lost something near 50 elo from playing with black in big part of your games instead of playing with white that is better for rating.

It seems that with fair rating list and wise choice to play only white you could get probably rating of 2450-2500 that is not enough for first place in the list.

Earlier I made a mistake when I did not consider that you get 1/4 out of 1/2 instead of 1/4 out of 1 but it did not change a lot.

More details:
I see that from 20.11.2024 you got 123 draws with white out of 411 games and 41 draws out of 540 games with black.
some of them were 1+1 but most of them are 1+0.
It seems that some bug in the search prevent me to count the number of 1+0 games and I did not like to count manually.

LeelaQueenOdds as black 1+0 (Reference) 2795.2

123 divided by 822 possible half points is 0.1496...
Assuming you get 15% against the bot then you should be 296 elo weaker that means rating of 2499.2
with the unfair calculation you get 123/4 points out of possible 123/2+(411-123) points or 123/(246+288*4)=0.0879 and it means something like 390 elo weaker than the bot and it means performance near 2405.2.

LeelaQueenOdds as white 1+0 (Reference) 2895.2
41 draws out of 540 is 41/1080 that is 0.03796
It means performance of more than 500 elo weaker than your opponent so with fair rating list you may get something like rating of 2390 instead of something near 2300 that you get with 41/4 out of 41/2+499 that is 41/(82+1996)=0.01973...

Uri. Thank you very much for your message, analysis and conclusions. For me it is very important to evaluate all this data, because it allows a double view in my mind: What was, what is and what could become. Knowing that I have launched myself into the bullring, without further predictions, it seems like a good result in this first approach to the breed bull (cow) LeelaQueenOdds. I would like to go back to bullfighting more, but I don't want to be attacked by the chubs. Having a reference of these is for me a good first point, but I would like more and better. I wonder: Applying it to everyone and equally in the top 10 table, would it put me closer or further away from leadership? The unknown Master and I, at what distance are we under the magnifying glass of general, impersonal and abstract norms, Fide? If I return, how many games and under what time controls should I play to earn more elo? Online parity is enough for me, or a good number of games, or having a good percentage, how many games should I play daily? Again Uri: Thank you. And to you Uri and the entire talchess group, I wish you a happy Thanksgiving.

[lkaufman]

To summerize the final point is that you probably lose something near 100 elo from the unfair rating list.
You also lost something near 50 elo from playing with black in big part of your games instead of playing with white that is better for rating.

It seems that with fair rating list and wise choice to play only white you could get probably rating of 2450-2500 that is not enough for first place in the list.

Earlier I made a mistake when I did not consider that you get 1/4 out of 1/2 instead of 1/4 out of 1 but it did not change a lot.

More details:
I see that from 20.11.2024 you got 123 draws with white out of 411 games and 41 draws out of 540 games with black.
some of them were 1+1 but most of them are 1+0.
It seems that some bug in the search prevent me to count the number of 1+0 games and I did not like to count manually.

LeelaQueenOdds as black 1+0 (Reference) 2795.2

123 divided by 822 possible half points is 0.1496...
Assuming you get 15% against the bot then you should be 296 elo weaker that means rating of 2499.2
with the unfair calculation you get 123/4 points out of possible 123/2+(411-123) points or 123/(246+288*4)=0.0879 and it means something like 390 elo weaker than the bot and it means performance near 2405.2.

LeelaQueenOdds as white 1+0 (Reference) 2895.2
41 draws out of 540 is 41/1080 that is 0.03796
It means performance of more than 500 elo weaker than your opponent so with fair rating list you may get something like rating of 2390 instead of something near 2300 that you get with 41/4 out of 41/2+499 that is 41/(82+1996)=0.01973...

Uri. Thank you very much for your message, analysis and conclusions. For me it is very important to evaluate all this data, because it allows a double view in my mind: What was, what is and what could become. Knowing that I have launched myself into the bullring, without further predictions, it seems like a good result in this first approach to the breed bull (cow) LeelaQueenOdds. I would like to go back to bullfighting more, but I don't want to be attacked by the chubs. Having a reference of these is for me a good first point, but I would like more and better. I wonder: Applying it to everyone and equally in the top 10 table, would it put me closer or further away from leadership? The unknown Master and I, at what distance are we under the magnifying glass of general, impersonal and abstract norms, Fide? If I return, how many games and under what time controls should I play to earn more elo? Online parity is enough for me, or a good number of games, or having a good percentage, how many games should I play daily? Again Uri: Thank you. And to you Uri and the entire talchess group, I wish you a happy Thanksgiving.

I agree with Uri, if draws counting fully instead of half, your rating might be almost one hundred higher. Of course anyone else who plays for draws would also be higher, but I don't know if anyone else does so. More generally, anyone else who scores below 50% at their chosen time limit would also benefit from the normal rules. If you play a lot of games, it is the percentage you score (at your chosen time control and color) that matters for rating, it doesn't matter much whether you play a hundred games or a thousand games unless your percentage improves with more games. As to the best time control to get a good rating, for most people, who play to win rather than to draw, I think they should play at a TC where they can usually win, but not too easily or consistently. But if you only play to draw, that is a different situation, where you should pick the TC where you can usually draw, but not always of course. Maybe for you 3 + 0 would work well, it is your favorite (along with 1+0) time control, you should be able to get a lot of draws I imagine, especially when you play White. Note that there are now quite a few top players (2800 to 3100 bullet) playing many 1+0 games with the bot, so they are playing under similar conditions to you, also getting only half the draws rated in effect, so your rating relative to those players should be quite fair. But none of them has a huge number of games yet like you do. In case you didn't know, I am "Hissha" on the list, just one place above you now. I really like increment, so I've settled on 1+10" (which counts as blitz) for this, where I can usually win but sometimes lose or draw, it's not so easy. At least I don't lose on time now. It doesn't show who I am, but I use the same handle on chess.com, where it shows my name and title.