Queen for Knight challenge

[chrisw]

In order to spur interest in the little-known LeelaQueenForKnight bot, I offered a $300 prize for the first FIDE titled player of known identity (NMs allowed if rated over 2600 blitz or Rapid on Lichess or chess.com) to win a 30 game blitz match from the bot. Initially it was restricted to 3'0" (or 2'1"), but since no one even came close in the first 3 days, despite many IMs and one GM trying, the max time limit has been raised to 3'1", which many consider to be comparable online to the standard over-the-board blitz time control of 3'2". Maybe soon we will see someone claim the prize. There are some restrictions, Leela gets White, the player must have an established online blitz or Rapid rating over 2500, the 30 games must be consecutive and played within a 16 hour window, and the prize won't be awarded if there is evidence of cheating (computer use). The title and rating restrictions are meant to weed out players who would almost certainly not be able to win the match without cheating; exceptions will be made for known strong players on request. In general, players below FM level can't even win a blitz match at full queen odds, so it probably takes GM blitz level to do so at queen for knight odds. But at 3'0" or 2'1" the best result so far was by IM Kacper Drozdowski, a really top level blitz player (3070 on chess.com), who scored 8 points in the 30 games.

It’s beyond me how bullet and blitz prove anything. It’s way too easy to make dumb errors of blindness at those time controls. Results are a fix. Please try again at 30 minutes or one hour per game.

Certainly it is more interesting to have longer TC games at smaller odds, but it is far more difficult to persuade titled players to play such games for modest prizes, and unknown amateurs with no chess reputation to risk are far more likely to cheat for money prizes. Also the risk of cheating rises with the time limit, or more precisely the risk of undetectable cheating rises; it's usually fairly obvious in blitz. Of course there are more and larger errors as the time limit gets faster, but in general I don't think there is much difference between a strong GM playing blitz at 3'1" than a random 2200 FIDE player playing 30'10". Just assume that whatever performance rating a bot gets at 3'1" will be several hundred elo lower at long time controls; with more data we can quantify that number more precisely.

Well, you can go to longer time controls without cash prizes. If people are interested in playing they will play, or is your objective just to maximise games played for some reason?

Obectives include increasing interest in the bots, determing their elo more accurately at the time controls preferred by strong players, and comparing the different handicaps. People are welcome/encouraged to play at longer time controls, but what is a competition without a prize? We might very well have a similar competition at a smaller handicap with a longer time limit, but for the chosen handicap the time limit was appropriate; the prize was won today by Super-GM Jeffery Xiong, after several other strong GMs and IMs failed. Although of course some games were won by the bot due to simple tactics, others were more interesting/impressive, won against multiple small errors. The next challenge will be a similar one but at odds of Two Knights, details to be announced soon.

Have you filtered the games for non-simple tactics and trapped the “interesting” ones? I’m sure a PDF or two or three of these “impressive” bot wins would interest. Especially the queen odds ones.

[lkaufman]

Have you filtered the games for non-simple tactics and trapped the “interesting” ones? I’m sure a PDF or two or three of these “impressive” bot wins would interest. Especially the queen odds ones.

I haven't had the time to do that so far, but "Chess-Network" regularly sifts thru these odds games and makes youtube videos on the most interesting ones, although almost all of his selections are either knight odds or rook odds so far. He does a great job of both selection and commentary. It's very difficult to prune non-simple tactics by automated methods, because many games may end with such tactics but only because the human was put under so much pressure that he used most of his time and was already unlikely to win anyway. Or one tactical error may still leave the human clearly winning, only to be gradually outplayed. In general, the human losses are partly due to being generally outplayed and partly due to tactical blunders, it's just a question of the proportion of each and the size of the blunders. When the odds reach full queen odds, there does usually have to be some fairly obvious blunder to lose, but not always.

[chrisw]

Have you filtered the games for non-simple tactics and trapped the “interesting” ones? I’m sure a PDF or two or three of these “impressive” bot wins would interest. Especially the queen odds ones.

I haven't had the time to do that so far, but "Chess-Network" regularly sifts thru these odds games and makes youtube videos on the most interesting ones, although almost all of his selections are either knight odds or rook odds so far. He does a great job of both selection and commentary. It's very difficult to prune non-simple tactics by automated methods, because many games may end with such tactics but only because the human was put under so much pressure that he used most of his time and was already unlikely to win anyway. Or one tactical error may still leave the human clearly winning, only to be gradually outplayed. In general, the human losses are partly due to being generally outplayed and partly due to tactical blunders, it's just a question of the proportion of each and the size of the blunders. When the odds reach full queen odds, there does usually have to be some fairly obvious blunder to lose, but not always.

Well, you said there were "more interesting and impressive games", but you can't show any. Pointless conversation from my POV, you make assertions but decline the source data. End of conversation, really.

[lucario6607]

Have you filtered the games for non-simple tactics and trapped the “interesting” ones? I’m sure a PDF or two or three of these “impressive” bot wins would interest. Especially the queen odds ones.

I haven't had the time to do that so far, but "Chess-Network" regularly sifts thru these odds games and makes youtube videos on the most interesting ones, although almost all of his selections are either knight odds or rook odds so far. He does a great job of both selection and commentary. It's very difficult to prune non-simple tactics by automated methods, because many games may end with such tactics but only because the human was put under so much pressure that he used most of his time and was already unlikely to win anyway. Or one tactical error may still leave the human clearly winning, only to be gradually outplayed. In general, the human losses are partly due to being generally outplayed and partly due to tactical blunders, it's just a question of the proportion of each and the size of the blunders. When the odds reach full queen odds, there does usually have to be some fairly obvious blunder to lose, but not always.

Well, you said there were "more interesting and impressive games", but you can't show any. Pointless conversation from my POV, you make assertions but decline the source data. End of conversation, really.

You weren’t contributing anyways, so thank you for seeing yourself out.

[chrisw]

Have you filtered the games for non-simple tactics and trapped the “interesting” ones? I’m sure a PDF or two or three of these “impressive” bot wins would interest. Especially the queen odds ones.

I haven't had the time to do that so far, but "Chess-Network" regularly sifts thru these odds games and makes youtube videos on the most interesting ones, although almost all of his selections are either knight odds or rook odds so far. He does a great job of both selection and commentary. It's very difficult to prune non-simple tactics by automated methods, because many games may end with such tactics but only because the human was put under so much pressure that he used most of his time and was already unlikely to win anyway. Or one tactical error may still leave the human clearly winning, only to be gradually outplayed. In general, the human losses are partly due to being generally outplayed and partly due to tactical blunders, it's just a question of the proportion of each and the size of the blunders. When the odds reach full queen odds, there does usually have to be some fairly obvious blunder to lose, but not always.

Well, you said there were "more interesting and impressive games", but you can't show any. Pointless conversation from my POV, you make assertions but decline the source data. End of conversation, really.

You weren’t contributing anyways, so thank you for seeing yourself out.

Pfff!! I’m the person who generated the opening books of thousands of used (by Larry, although he may have updated by now) for playing odds games. I’m also a chess programmer since 1980, hence entitled to comment on matters relevant. Learn some manners and do your research before making dumb comments.

[lucario6607]

Have you filtered the games for non-simple tactics and trapped the “interesting” ones? I’m sure a PDF or two or three of these “impressive” bot wins would interest. Especially the queen odds ones.

I haven't had the time to do that so far, but "Chess-Network" regularly sifts thru these odds games and makes youtube videos on the most interesting ones, although almost all of his selections are either knight odds or rook odds so far. He does a great job of both selection and commentary. It's very difficult to prune non-simple tactics by automated methods, because many games may end with such tactics but only because the human was put under so much pressure that he used most of his time and was already unlikely to win anyway. Or one tactical error may still leave the human clearly winning, only to be gradually outplayed. In general, the human losses are partly due to being generally outplayed and partly due to tactical blunders, it's just a question of the proportion of each and the size of the blunders. When the odds reach full queen odds, there does usually have to be some fairly obvious blunder to lose, but not always.

Well, you said there were "more interesting and impressive games", but you can't show any. Pointless conversation from my POV, you make assertions but decline the source data. End of conversation, really.

You weren’t contributing anyways, so thank you for seeing yourself out.

Pfff!! I’m the person who generated the opening books of thousands of used (by Larry, although he may have updated by now) for playing odds games. I’m also a chess programmer since 1980, hence entitled to comment on matters relevant. Learn some manners and do your research before making dumb comments.

Arguably, I would say I am more qualified than you for this topic.

[chrisw]

Have you filtered the games for non-simple tactics and trapped the “interesting” ones? I’m sure a PDF or two or three of these “impressive” bot wins would interest. Especially the queen odds ones.

I haven't had the time to do that so far, but "Chess-Network" regularly sifts thru these odds games and makes youtube videos on the most interesting ones, although almost all of his selections are either knight odds or rook odds so far. He does a great job of both selection and commentary. It's very difficult to prune non-simple tactics by automated methods, because many games may end with such tactics but only because the human was put under so much pressure that he used most of his time and was already unlikely to win anyway. Or one tactical error may still leave the human clearly winning, only to be gradually outplayed. In general, the human losses are partly due to being generally outplayed and partly due to tactical blunders, it's just a question of the proportion of each and the size of the blunders. When the odds reach full queen odds, there does usually have to be some fairly obvious blunder to lose, but not always.

Well, you said there were "more interesting and impressive games", but you can't show any. Pointless conversation from my POV, you make assertions but decline the source data. End of conversation, really.

You weren’t contributing anyways, so thank you for seeing yourself out.

Pfff!! I’m the person who generated the opening books of thousands of used (by Larry, although he may have updated by now) for playing odds games. I’m also a chess programmer since 1980, hence entitled to comment on matters relevant. Learn some manners and do your research before making dumb comments.

Arguably, I would say I am more qualified than you for this topic.

Which is of course why you made technical, knowledgeable and/or chess specific posts into this thread. Not.
I just see personal and rude BS. Have a nice day. Bye.

[towforce]

I haven't had the time to do that so far, but "Chess-Network" regularly sifts thru these odds games and makes youtube videos on the most interesting ones, although almost all of his selections are either knight odds or rook odds so far. He does a great job of both selection and commentary. It's very difficult to prune non-simple tactics by automated methods, because many games may end with such tactics but only because the human was put under so much pressure that he used most of his time and was already unlikely to win anyway. Or one tactical error may still leave the human clearly winning, only to be gradually outplayed. In general, the human losses are partly due to being generally outplayed and partly due to tactical blunders, it's just a question of the proportion of each and the size of the blunders. When the odds reach full queen odds, there does usually have to be some fairly obvious blunder to lose, but not always.

Most of the time, a human blunder is a result of being in a difficult position. The whole idea of these odds challenge engines is to get the human into difficult positions they cannot accurately cope with.

With good play, a player would win with queen for knight odds, so the engine needs to get the human to blunder.

It follows that these engines are designed to induce human blunders, and cannot win, or even draw, without one.