I just ran it against lichess’ stockfish 14.1 nnue infinite analysis and with a minute on each step it ended with a draw so it’s smth around 3300-3400
What is the mittens bots real elo?
Sort:
Jan 1, 2023
0
#23
Mittens is cheating lol. I played bullet against her and flagged her.
Did anyone win against Mitten Bot? And if u did, how??? I think the mitten bot's real elo is 2500+
Koinichiwa wrote:
anything over 2250 for sure
More like anything over 3250
Jan 1, 2023
0
#28
I played agaisnt it and it was a Najdorf. It managed to destroy me after I messed something up
Jan 1, 2023
0
#29
I agree with the ~stockfish level mittens elo. On my mobile chess.com, mittens thinks for times comparable to the 3200 engine, which makes me suspec that...
But why would chess.com upload a bot stronger than its "max"
Jan 1, 2023
0
#30
Jan 1, 2023
0
#31
Flocrow wrote:
I just turned on Stockfish and played Mitten, I played the best move Stockfish suggested and I ended in a draw
I think the best way Stockfish can beat him is Ruy Lopez. Stable advantage which this bot hates.
Jan 1, 2023
0
#33
I would say its rating is approximately 2600, considering it took Stockfish 50 moves to win.
Jan 1, 2023
0
#34
People are throwing around accuracy scores when playing Mittens as if they would mean a thing. They are measured against whatever particular engine and depth chesscom is using in review, so essentially you are using a weaker engine to judge a stronger one. For example, I just beat Mittens using Lc0 v28 (leela) and analysis showed some moves as inaccuracies that clearly weren't, because lc0 has better judgement than the analysis engine. Also in the endgame, it didn't see several moves as best moves even though they were objectively the best (according to tablebase).
Anyway, for your enjoyment, lc0 v28 vs Mittens. It's also a very instructive knight vs knigh endgame, where white had to navigate a long sequence of nontrivial only moves to maintain a winning advantage:
https://www.chess.com/analysis/game/computer/29331847?tab=review
I resigned after it put me in a decently bad situation. I fear no man, but that thing, it scares me.
Jan 1, 2023
0
#36
The people who drew or beat it used 1|0 games... and yeah, with zero warm up I beat mittins in 22 moves when the time control is 1|0... and I almost always lose my first 2 or 3 games when I play actual humans so the fact that I won with no warmup...
-
-
I played it with no time control and I resigned on move 34 and it has 97% accuracy
-
Thank goodness it's not just me. I thought I was the only loser consistently losing to a level 1 cat.
Jan 1, 2023
0
#39
Stockfish 13, depth = 30 with opening book vs Mittens
-
-
Chess.com analysis at depth = 30.
Jan 1, 2023
0
#40
So what's the purpose of all of this now? To figure out what engine with what configuration is behind it? The analysis tool is not helpful for that, it only compares Mittens to whatever engine is used for analysis. To determine what's behind it, or even estimate its strength, one would have to download a lot of its games against strong engines and analyse the overlap with various other engines offline, like what people did with Niemann games... and then again, it's unlikely to lead to 100% coincidences as it's probably based on a highly customized version. Or maybe chesscom made a deal with Hans Niemann... they let him back on the site and gave him permission to use engines, as long as he pretends to be a bot.
On the other hand, I figured this can generate very instructive games.
I have been researching Mittens. She appears to be Stockfish (faster) with depth 18. That is the same engine used for suggestions and hints. I managed to beat her with Stockfish (NNUE) with depth 40. My accuracy was 98.7 and hers was 81.7. She seems to draw with depth 18 and suggestions, leading me to the conclusion. Some have said that it cannot be stockfish due to copyright issues, but stockfish is open source. Therefore, Mittens' elo is around 3700. I hope you find this helpful.