Amazing results of LeelaKnightOdds-Dev

[Uri Blass]

after additional losses in game 4 and game 5 I got a draw in game 6
Of course I knew that lc0 does not know my rating and is happy with a draw but it was relatively my best game with knight odds.

Maybe later I will try to win with a winning position but I am afraid I need more time for it to have practical chances and 5+3 is not enough.

https://lichess.org/0QdPn8LhnVov

[Father]

after additional losses in game 4 and game 5 I got a draw in game 6
Of course I knew that lc0 does not know my rating and is happy with a draw but it was relatively my best game with knight odds.

Maybe later I will try to win with a winning position but I am afraid I need more time for it to have practical chances and 5+3 is not enough.

https://lichess.org/0QdPn8LhnVov

Congratulations Uri. A great match, which under a simple rule of three calculation, would yield a result in one hundred and fifty-nine games, would give just over thirteen points, which would represent a performance greater than 200% than that of the Goliaths. The shield and the sword were the strategic steps of the giant, the sling and the stone are the weapons of the weakest. Choosing the sling and the stone were David's strategic procedures, launching them against the giant Goliath was the tactical development. Congratulations.

[lkaufman]

after additional losses in game 4 and game 5 I got a draw in game 6
Of course I knew that lc0 does not know my rating and is happy with a draw but it was relatively my best game with knight odds.

Maybe later I will try to win with a winning position but I am afraid I need more time for it to have practical chances and 5+3 is not enough.

https://lichess.org/0QdPn8LhnVov

The results in Rapid games (so slower than your games) since the Sep 28 upgrade, against players rated over 2500 only, are 54 wins for the bot, 5 draws, and ZERO losses. Performance rating in Rapid about 3060. No one over 1800 has won a single game of Rapid since the upgrade. Some players rated below that or unrated have "wins", but they invariably show 98 or 99% accuracy per Stockfish which is extremely implausible for players below 1800 Rapid on Lichess, so given that no one above 1800 has won I would assume these were all cheaters using Stockfish. It seems that only low rated and especially unrated players cheat, the higher rated ones have presumably played a lot without getting banned and would have more to lose by being banned, with nothing to gain since the games are unrated. But weak players apparently want to "brag" about "beating" a bot, even at knight odds and using an engine!!

[Father]

after additional losses in game 4 and game 5 I got a draw in game 6
Of course I knew that lc0 does not know my rating and is happy with a draw but it was relatively my best game with knight odds.

Maybe later I will try to win with a winning position but I am afraid I need more time for it to have practical chances and 5+3 is not enough.

https://lichess.org/0QdPn8LhnVov

The results in Rapid games (so slower than your games) since the Sep 28 upgrade, against players rated over 2500 only, are 54 wins for the bot, 5 draws, and ZERO losses. Performance rating in Rapid about 3060. No one over 1800 has won a single game of Rapid since the upgrade. Some players rated below that or unrated have "wins", but they invariably show 98 or 99% accuracy per Stockfish which is extremely implausible for players below 1800 Rapid on Lichess, so given that no one above 1800 has won I would assume these were all cheaters using Stockfish. It seems that only low rated and especially unrated players cheat, the higher rated ones have presumably played a lot without getting banned and would have more to lose by being banned, with nothing to gain since the games are unrated. But weak players apparently want to "brag" about "beating" a bot, even at knight odds and using an engine!!

… these results code is very interesting... in the near future, I will try to win one or more games, I will probably have to use time control with increment and take my time to think in greater depth. I don't know what my playing strength would be on the Fide scale if I were to compete in open tournaments regularly in a six-month interval. The problem is that I really have no interest in having it. In the past, many years ago, I faced the computers with a time of one hour plus 15 seconds per move per game and it was very interesting. The matches were made in ten games alternating colors. As a young man I participated in very few tournaments, I was champion of my university, and I won an open in which the national youth champion competed, playing the last game a dragon defense of the Sicilian leading the black chips. That's when computers arrived and I never played any tournaments again...

[Uri Blass]

after additional losses in game 4 and game 5 I got a draw in game 6
Of course I knew that lc0 does not know my rating and is happy with a draw but it was relatively my best game with knight odds.

Maybe later I will try to win with a winning position but I am afraid I need more time for it to have practical chances and 5+3 is not enough.

https://lichess.org/0QdPn8LhnVov

The results in Rapid games (so slower than your games) since the Sep 28 upgrade, against players rated over 2500 only, are 54 wins for the bot, 5 draws, and ZERO losses. Performance rating in Rapid about 3060. No one over 1800 has won a single game of Rapid since the upgrade. Some players rated below that or unrated have "wins", but they invariably show 98 or 99% accuracy per Stockfish which is extremely implausible for players below 1800 Rapid on Lichess, so given that no one above 1800 has won I would assume these were all cheaters using Stockfish. It seems that only low rated and especially unrated players cheat, the higher rated ones have presumably played a lot without getting banned and would have more to lose by being banned, with nothing to gain since the games are unrated. But weak players apparently want to "brag" about "beating" a bot, even at knight odds and using an engine!!

It seems that the only player above rating 2000 who won against it in lichess is in blitz and it is AdmiralHabibi who got 1 win out of 146 games and lost in 139 games(he stopped playing against lc0 after the win).

[lkaufman]

after additional losses in game 4 and game 5 I got a draw in game 6
Of course I knew that lc0 does not know my rating and is happy with a draw but it was relatively my best game with knight odds.

Maybe later I will try to win with a winning position but I am afraid I need more time for it to have practical chances and 5+3 is not enough.

https://lichess.org/0QdPn8LhnVov

The results in Rapid games (so slower than your games) since the Sep 28 upgrade, against players rated over 2500 only, are 54 wins for the bot, 5 draws, and ZERO losses. Performance rating in Rapid about 3060. No one over 1800 has won a single game of Rapid since the upgrade. Some players rated below that or unrated have "wins", but they invariably show 98 or 99% accuracy per Stockfish which is extremely implausible for players below 1800 Rapid on Lichess, so given that no one above 1800 has won I would assume these were all cheaters using Stockfish. It seems that only low rated and especially unrated players cheat, the higher rated ones have presumably played a lot without getting banned and would have more to lose by being banned, with nothing to gain since the games are unrated. But weak players apparently want to "brag" about "beating" a bot, even at knight odds and using an engine!!

It seems that the only player above rating 2000 who won against it in lichess is in blitz and it is AdmiralHabibi who got 1 win out of 146 games and lost in 139 games(he stopped playing against lc0 after the win).

That player "AdmiralHabibi" is an IM with a FIDE rating of 2420! All of his games were 3'2", no games without increment. I don't know if even Magnus Carlsen could achieve results like this against a similar field playing normal chess with no handicap. This suggests, but does not prove, that the bot could give Magnus Carlsen knight odds in blitz, maybe not yet in Rapid.

[Uri Blass]

I see that Leela lost in rapid against Sistina but it seems that Sistina lost 89 games and drew one game before winning.

[Uri Blass]

I could win yesterday in my only try against LeelaQueenForKnight
Here is the game

https://lichess.org/PzKxEehQ

Moshe Nahmias(fide rating 1572) had less luck in that time control and lost 3 games(at least he could win against the queen odds version of leela with the white pieces(did not try with black)

[lkaufman]

I could win yesterday in my only try against LeelaQueenForKnight
Here is the game

https://lichess.org/PzKxEehQ

Moshe Nahmias(fide rating 1572) had less luck in that time control and lost 3 games(at least he could win against the queen odds version of leela with the white pieces(did not try with black)

That's actually an accomplishment for you. It is rather unusual for anyone under LiChess 2200 to win at queen for knight, esoecially with the engine playing White. Performance rating is in the ballpark of 2400 as White I think. The knight-odds trained bot plays queen for knight even better than the current queen for knight bot.

[lkaufman]

I see that Leela lost in rapid against Sistina but it seems that Sistina lost 89 games and drew one game before winning.

The regular LeelaKnightOdds bot started using the LeelaKnightOdds-dev net (trained on knight odds) on the afternoon of Oct 17, so both both use it now, and the dev bot will be retired from general use at the end of the month. During the last three days, prior to the games you mention, the two bots together using the new net won over 700 CONSECUTIVE games, not even giving up a draw! The majority of the opponents were rated over 2000, with 65 of the 700 wins against players rated over 2500! Tonite it lost a 15'5" Game to 2739 rated "Chess-Network", who had lost his first two games with it at that tc.

Grandmaster Awonder Liang, who is currently in second place halfway thru the U.S. Championship, will play a Rapid knight odds match with the bot soon after the championship. He talked about it in his post-game interview today after a spectacular win, suggesting that his preparation for the knight odds match may have inspired his sacrificial play. I expect a close match.