Could Stockfish win at knight handicap against a GM ?

[JJJ]

So finally, maybe we re gonna know a little better the true rating of Komodo thanks to many handicap match played by everyone.

I guess for each handicap Komodo is gonna have a different name ? Like Komodo-norook Komodo-noknight and each name having a different rating ?

[lkaufman]

So finally, maybe we re gonna know a little better the true rating of Komodo thanks to many handicap match played by everyone.

I guess for each handicap Komodo is gonna have a different name ? Like Komodo-norook Komodo-noknight and each name having a different rating ?

Well, that's the way I would do it, but I don't yet know what chess.com will do. For now, it's just "playkomodo" playing at different handicaps, and I'll keep the stats. This Saturday from about 4 to 7 pm Eastern time Komodo will play some handicap games on chess.com with my son Raymond (IM). Nothing official but he'll play his best, we want to learn something from this. Perhaps knight odds in blitz and smaller handicaps in rapid (15' + 10" probably). Anyone can watch. We're still trying to make sure it all works smoothly; there were some bugs with this automated play but I think they are fixed now.

[JJJ]

Thanks you Larry.

[Ovyron]

Hi Larry. Have you considered "Opening Position Handicaps"?

The idea is to start the game against the human from the Opening Position, and allow Komodo to create a Handicap of his own as the game progresses.

A way to achieve this is to record, from the position with the handicap (say, the one without a Knight present) how many centipawns Komodo thinks this one has an advantage. Then you start the game against the human from starting position with 2 Komodos, one that is some insane thing that plays anything (it could be a random mover, or Komodo at Depth 1, or something else), while the other makes sure that this Komodo doesn't make a move that passes the Handicap Margin (if the margin is -2.00 and the move is -2.01 you don't let the insane thing make it).

Once this margin is reached (or closely matched; maybe -2.01 is good enough for a margin of -2.00...) Komodo would play at his best, from the resulting position, just like from the starting position without a knight.

So, instead of "missing pawns" or "missing pieces", you'd have some "centipawns score" handicaps, where Komodo will play dumbly until it is reached, but from the actual opening position of chess instead of one that looks as a variant.

[lkaufman]

Hi Larry. Have you considered "Opening Position Handicaps"?

The idea is to start the game against the human from the Opening Position, and allow Komodo to create a Handicap of his own as the game progresses.

A way to achieve this is to record, from the position with the handicap (say, the one without a Knight present) how many centipawns Komodo thinks this one has an advantage. Then you start the game against the human from starting position with 2 Komodos, one that is some insane thing that plays anything (it could be a random mover, or Komodo at Depth 1, or something else), while the other makes sure that this Komodo doesn't make a move that passes the Handicap Margin (if the margin is -2.00 and the move is -2.01 you don't let the insane thing make it).

Once this margin is reached (or closely matched; maybe -2.01 is good enough for a margin of -2.00...) Komodo would play at his best, from the resulting position, just like from the starting position without a knight.

So, instead of "missing pawns" or "missing pieces", you'd have some "centipawns score" handicaps, where Komodo will play dumbly until it is reached, but from the actual opening position of chess instead of one that looks as a variant.

Well, that is interesting. There are some practical problems though. For a chess.com show, we need some specific start position. I suppose we could do what you propose before the show, but that's a bit unsatisfying. Another issue is that computer evals (whether Komodo, SF or Houdini doesn't matter) aren't good predictors of handicap values, it seems. Basically material is easier for the human to exploit than dynamic factors like time and king safety. We've seen this again and again. Humans never win and rarely even draw when the handicap is mostly dynamic, but give them a knight and they win pretty easily.

The closest we've come to your idea so far is the match with GM Perelshteyn, where I chose four awful openings (like e4 e6 d4 Ke7? or d4 g5?) that produced fairly similar evals. One other point to keep in mind is that almost all of the handicaps we've tried could arise by a legal sequence from the start position, although of course they would be silly sequences. There is some appeal to your idea which only requres the engine to make bad moves. Also your idea avoids preparation. I'll have to think more about all of this. Suggestions are welcome.

[JJJ]

If Komodo is allowed to play on chess.com with multiple handicap will you test many different contempt to know which one is best for each handicap ?

Also, how would you compare the elo of Komodo without handicap with Komodo handicap ? I mean contempt might change elo.

[lkaufman]

If Komodo is allowed to play on chess.com with multiple handicap will you test many different contempt to know which one is best for each handicap ?

Also, how would you compare the elo of Komodo without handicap with Komodo handicap ? I mean contempt might change elo.

If it's practical, I'll try multiple contempt values. I have a rough idea of what's best for each handicap now, but I could be off somewhat from optimum. Of course contempt can change elo, but I think it's fair to measure elo with a realistic contempt value. After all Magnus Carlsen knows how much contempt to have for each opponent he plays, and his rating is based on that knowledge, so why shouldn't the engine have this knowledge too if it is to be rated against humans? This is especially true for handicap games, because the human knows to avoid complications and simplify, so the engine needs to know to do the opposite.

[Ovyron]

The closest we've come to your idea so far is the match with GM Perelshteyn, where I chose four awful openings (like e4 e6 d4 Ke7? or d4 g5?) that produced fairly similar evals.

I see, this seems to have all the wanted benefits, except the human is forced some moves.

My suggestion would be to expand on this, and create some Awful Opening book that reaches those awful positions no matter what the human plays. I believe the book could be given to the human so they prepare at home, and might even choose what awful openings they'd play (on the board, so they play the continuation that they know the book would follow.)

The highest problem would be making sure that the human doesn't play into some hole of the book, where an unexpected continuation would leave Komodo in a good position at full strength, but I believe it'd be the human's fault if they didn't do their homework and don't really know for what advantageous position to reach. But this seems more satisfactory than some random pieces or pawns missing.

The main part I don't like about current handicaps is that both sides have to collude to make one side lose some material before game start, though I don't mind if it's accompanied by interesting commentary and reactions like the ones by Daniel Rensch when he played Komodo at handicaps on Youtube (which was good stuff, the opening position or how it was reached didn't really matter thanks to the presentation.)

Have you considered having someone be a commentator of the handicap games live as they happen while streaming in a place like Youtube or Twitch?

[lkaufman]

The closest we've come to your idea so far is the match with GM Perelshteyn, where I chose four awful openings (like e4 e6 d4 Ke7? or d4 g5?) that produced fairly similar evals.

I see, this seems to have all the wanted benefits, except the human is forced some moves.

My suggestion would be to expand on this, and create some Awful Opening book that reaches those awful positions no matter what the human plays. I believe the book could be given to the human so they prepare at home, and might even choose what awful openings they'd play (on the board, so they play the continuation that they know the book would follow.)

The highest problem would be making sure that the human doesn't play into some hole of the book, where an unexpected continuation would leave Komodo in a good position at full strength, but I believe it'd be the human's fault if they didn't do their homework and don't really know for what advantageous position to reach. But this seems more satisfactory than some random pieces or pawns missing.

The main part I don't like about current handicaps is that both sides have to collude to make one side lose some material before game start, though I don't mind if it's accompanied by interesting commentary and reactions like the ones by Daniel Rensch when he played Komodo at handicaps on Youtube (which was good stuff, the opening position or how it was reached didn't really matter thanks to the presentation.)

Have you considered having someone be a commentator of the handicap games live as they happen while streaming in a place like Youtube or Twitch?

The chess.com matches are broadcast on chess.com/tv and Twitch. Some are commented by the players themselves, others by outside commentators. In the Perelshteyn match he was given the four bad opening positions well in advance of the match. One problem with this approach is that in general pawn-down positions are not bad enough without some added bad moves, while piece down positions require some compensation which either means a pawn or some bad moves by the human. Also it's hard to lose the Exchange in the opening without silly moves. But these are not fatal problems.

[Ovyron]

The chess.com matches are broadcast on chess.com/tv and Twitch. Some are commented by the players themselves, others by outside commentators.

Thanks, I had no idea about this.

Since this is content that interests me I suggest reuploading those matches with commentary to Youtube, perhaps in chess.com's channel or another one. I doubt I'm the only one enjoying those matches but that had no idea there were twitch streams.