'Mittens ChessBot' (Chess.com) Stuns Chessworld

[lkaufman]

I watched a video of GM Nakamura beating this bot with pawn odds in his favor:


He had several difficult decisions to make during the game and probably made some inaccuracies, but in the end he won.
What most impressed me is that a lot of time he was calculating, he wasn't looking at the board.

I don't have any inside information specifically on "Mittens" (if I did it would be confidential), but the regular engine bots where you just set the level from 1 to 25 are just a recent Komodo (not NNUE), limited by time (or nodes) on level 25, and limited on the lower levels by nodes and variety (randomness) as in the elo levels on regular Komodo or Komodo Dragon. Level 25 is stated as being Elo 3200, which is probably reasonable as a Rapid rating for regular Komodo taking about a second per move. I think Hikaru did win one out of many games against the top bot level of Komodo a few years ago (when it was a bit weaker than now), and I would expect that pawn odds would be reasonably competitive now for that bot with Hikaru playing Rapid (remember Dragon with many cores beat him 6.5 to 1.5 in Rapid giving two non-edge pawns). So my question is this: is there any reason to think that "Mittens" is any stronger than chess.com Engine 25 (aka Komodo single thread no NNUE taking a second per move or so)? I would think that the simplest way to make "Mittens" would just be to use full strength Komodo and change some of the parameter settings to achieve some desired properties, presumably weakening it slightly overall. Is there any evidence that is incompatible with this hypothesis?

[towforce]

Mittens got Crushed by Stockfish 15.1 in the Stockfish Channel. Is Mittens Komodo Dragon? I just played Mittens and lasted 53 moves.

I have read that Chess.com bots are written in JavaScript - has that changed (web assembly (WASM) would be a better choice IMO)? It seems unlikely that the Komodo Dragon team would transpile their product into JavaScript for Chess.com's benefit.

Looks as though Daniel71 got this right and I got it wrong!

In a recent post in this thread, Larry Kaufman has revealed that many of chess.com's bots are actually Komodo Dragon! I literally had no idea, so the suggestion sounded crazy!

We have been told later in this thread that the bots are WASM based, so not JavaScript. Hence, we now have a good idea what's going on with most of the chess.com bots: they are Dragon's move selection code recompiled as WASM.

This also means that there's a good chance that Mittens is also a version of Komodo.

Larry Kaufman has asked if Mittens might also be (if I may paraphrase what he said) a non NNUE Dragon with some parameters amended. My answer: "no": I am personally nowhere near a strong enough player to judge how good Mittens is, but the people who are seem to think that it's a big jump from previous chess.com bots: this screams "NNUE" to me.

So once again, I'm amending my guess as to what Mittens is: I'm now going to say that it's Komodo Dragon with NNUE (whereas the previous chess.com bots didn't have NNUE), and it's a WASM application that runs in the user's browser. I'm guessing that the NNUE size is 20-40 mb.

This still leaves a big mystery, though: if the program is running in the user's browser, why did chess.com decide to weaken it's strength for a few days? That would only make sense if it was a server program?

[lkaufman]

Mittens got Crushed by Stockfish 15.1 in the Stockfish Channel. Is Mittens Komodo Dragon? I just played Mittens and lasted 53 moves.

I have read that Chess.com bots are written in JavaScript - has that changed (web assembly (WASM) would be a better choice IMO)? It seems unlikely that the Komodo Dragon team would transpile their product into JavaScript for Chess.com's benefit.

Looks as though Daniel71 got this right and I got it wrong!

In a recent post in this thread, Larry Kaufman has revealed that many of chess.com's bots are actually Komodo Dragon! I literally had no idea, so the suggestion sounded crazy!

We have been told later in this thread that the bots are WASM based, so not JavaScript. Hence, we now have a good idea what's going on with most of the chess.com bots: they are Dragon's move selection code recompiled as WASM.

This also means that there's a good chance that Mittens is also a version of Komodo.

Larry Kaufman has asked if Mittens might also be (if I may paraphrase what he said) a non NNUE Dragon with some parameters amended. My answer: "no": I am personally nowhere near a strong enough player to judge how good Mittens is, but the people who are seem to think that it's a big jump from previous chess.com bots: this screams "NNUE" to me.

So once again, I'm amending my guess as to what Mittens is: I'm now going to say that it's Komodo Dragon with NNUE (whereas the previous chess.com bots didn't have NNUE), and it's a WASM application that runs in the user's browser. I'm guessing that the NNUE size is 20-40 mb.

This still leaves a big mystery, though: if the program is running in the user's browser, why did chess.com decide to weaken it's strength for a few days? That would only make sense if it was a server program?

It may be much stronger than the named bots, which are node-limited, but is it stronger than unlimited skill 25, which is roughly Dragon 3.1 without NNUE? Results I have seen don’t support that claim.

[Jjaw]

If you have questions about mittens chess engine , download it & play against it ... I did ... https://workupload.com/file/gCAw6Eps5nd

[M ANSARI]

I don't get the big deal about this Mittens engine. I mean the cat pic looks cute, but how is this engine any different than the other hundreds of other engines that are not up to latest SF strength? Is there a specific reason why this engine is making headlines? It just seems like a weaker engine ... which of course would still do really well against any human!

[carldaman]

If you have questions about mittens chess engine , download it & play against it ... I did ... https://workupload.com/file/gCAw6Eps5nd

If Mittens can be freely downloaded, then maybe someone can run the similarity tool on it to see which engine it resembles the most.

[towforce]

I don't get the big deal about this Mittens engine. I mean the cat pic looks cute, but how is this engine any different than the other hundreds of other engines that are not up to latest SF strength? Is there a specific reason why this engine is making headlines? It just seems like a weaker engine ... which of course would still do really well against any human!

If you wish to play a strong engine, you either have to install one on your device, or play against a bot which is known to be a strong engine at Lichess. Now it looks as though chess.com is providing a really strong bot opponent, which is convenient to say the least.

[lkaufman]

I don't get the big deal about this Mittens engine. I mean the cat pic looks cute, but how is this engine any different than the other hundreds of other engines that are not up to latest SF strength? Is there a specific reason why this engine is making headlines? It just seems like a weaker engine ... which of course would still do really well against any human!

If you wish to play a strong engine, you either have to install one on your device, or play against a bot which is known to be a strong engine at Lichess. Now it looks as though chess.com is providing a really strong bot opponent, which is convenient to say the least.

Well, they already had one, namely engine set to maximum skill level 25, which is essentially Dragon 3.1 on 1 cpu without NNUE, so it should be pretty much equal to or a bit stronger than the last Komodo before Dragon, namely 14.1, which is 3448 on CCRL blitz. That is on an old I7; perhaps one of you can estimate how much slower it runs with WASM and can extrapolate downware the rating based on that; I imagine something like 3300. Surely there is no evidence that Mittens is appreciably stronger than 3300 CCRL blitz, is there? Aside from cuteness and a name, Mittens does have one other arguable advantage, it moves in a second or so instead of using all available time based on the time limit. This should reduce its blitz rating by a hundred elo or so, so perhaps I should ask whether there is any evidence that it plays blitz better than 3200 CCRL?

[towforce]

I don't get the big deal about this Mittens engine. I mean the cat pic looks cute, but how is this engine any different than the other hundreds of other engines that are not up to latest SF strength? Is there a specific reason why this engine is making headlines? It just seems like a weaker engine ... which of course would still do really well against any human!

If you wish to play a strong engine, you either have to install one on your device, or play against a bot which is known to be a strong engine at Lichess. Now it looks as though chess.com is providing a really strong bot opponent, which is convenient to say the least.

Well, they already had one, namely engine set to maximum skill level 25, which is essentially Dragon 3.1 on 1 cpu without NNUE, so it should be pretty much equal to or a bit stronger than the last Komodo before Dragon, namely 14.1, which is 3448 on CCRL blitz. That is on an old I7; perhaps one of you can estimate how much slower it runs with WASM and can extrapolate downware the rating based on that; I imagine something like 3300. Surely there is no evidence that Mittens is appreciably stronger than 3300 CCRL blitz, is there? Aside from cuteness and a name, Mittens does have one other arguable advantage, it moves in a second or so instead of using all available time based on the time limit. This should reduce its blitz rating by a hundred elo or so, so perhaps I should ask whether there is any evidence that it plays blitz better than 3200 CCRL?

I'll quickly give my own answer: I don't know - sorry.

[CornfedForever]

Someone asked 'what the big deal' is with Mittens. I think...basically that it talks trash as is beats you. It looked that way in the...think it was a Hikaru video I saw.
And...cats are cool.