Can Stockfish 16 Beat Stockfish 15?

[@JoeKempsey]

Hi Everyone! Giving a quick update for the next video. I wanted to make a video testing the Elephant Gambit (1.e4 e5 2.Nf3 d5 +0.92 eval) but Stockfish 16 (White) outplayed Stockfish 15 fairly easily. I attached PGN of this game in description if you're interested :) So tomorrow I'm going to try the slightly less dubious Albin Counter-Gambit (1.d4 d5 2.c4 e5 +0.70 eval), giving Stockfish 16 White and Stockfish 15 Black. The idea is to give Stockfish 16 a slight advantage and see if it ends up being enough to take down the slightly weaker Stockfish 15! Video should be up a bit later in the week. Cheers to all!!

[@Sack_Zement]

Yeah. That was to be expected. 80 ELO more are not so significant at a level like that. Most games will be draws.

[@Engine448]

its only 47 elo more

[@Sack_Zement]

Ok

[@shambhav9534]

I think you should try higher version gaps. Something like Sf11 would be nice. No way one version higher version is going to beat its past self. Any good engine can draw Stockfish with 10 seconds per move.

[@JoeKempsey]

Hi Shambhav! Thank you for your comment. Will do more non chess.com engine battles for a bit since the new chess.com bots aren’t too great (we have to wait some time till the Benji bot reaches max strength) Gonna test some openings that are considered slightly dubious (0.50 to 1.00 eval) and see if Stockfish 16 can pull away with advantage and win against SF15!

[@kuei4604]

There's no way for stockfish 16 to defeat stockfish 15 in a very single game. I think the former can only win like 2-3 games in 100 games and draw the other 97-98 games against the latter. (More probable with white pieces of course.)

[@dsaez7024]

Greetings. I already tried to explain this on other channels, but ignorance is bold, so it was not well received. You see, the starting position is a drawn position, and this is something that the newer modules are capable of displaying. Therefore, to really notice the differences between modern programs, it is necessary to utilize unbalanced positions, in such a way that the strongest engine wins with a certain advantage and draws with a certain disadvantage, playing the same position, both with white and black. This is what is known as game pairs, since with the traditional method, the strongest machines, eg stockfish, leela and dragon, would draw 100% of the time. The reason this happens is because the new chess engines are so robust and low-erroneous that they will always convert a drawn position into an effective draw, namely the default starting position. Any serious testing or relevant chess engine competition relies on game pairs to determine the difference in strength and produce a statistically significant result. For this, multiple series are organized starting from unbalanced, irregular and asymmetrical positions. That's where all the rating differences for the newer engines come from, because otherwise they would draw every game

[@JoeKempsey]

Hello dsaez, thank you very much for your comment This video is just to showcase how tough modern-day engines are to crack with all things balanced from the start. Concluding that at surface level, not much seems to have changed between Stockfish 15 and 16; both versions are monsters! I will certainly be doing handicap matches between these two engines, don’t worry :) And will take your points into account. Thinking of having SF 15 play some dubious gambits as the black pieces against stockfish. Or maybe pawn odds!

[@dsaez7024]

@@JoeKempsey A clear example would be facing two chess engines in a king ending with two pawns, a completely dead and drawn position. Sure, I can also draw stockfish, starting from a position with very few pieces on the board. Would it be fair to deduce that Stockfish 8 and 16 have the same strength because they draw in a position with 2 pawns on the board, without the possibility of making any progress? It turns out that the most advanced engines are so efficient that even the starting position is such a trivial, dry, static and plank position. For this reason, the development of future versions is no longer measured from the initial position. Pawn odds is excessive. My recommendation is that you use openings with an output between 0.75 and 1.75 advantage for either party. Those extra elo points involve Stockfish 16 being able to turn handicaps into draws and advantages into wins at a higher rate than Stockfish 15. The default starting position in chess has already exhausted its functionality among elite engines

[@JoeKempsey]

@@dsaez7024 Noted. Thank you for your thorough comments!!

[@Zatchurz]

Exactly what i was going to point out, HOWEVER you can see the difference in strength of the engines by the play, clearly SF16 had the advantage in the game above, but then again SF16 was white. We would need to see SF16 vs SF15 with flipped colors, but judging from the endgame position, SF16 had more of an advantage than white would get in a SF16 vs SF16 match. As I have mentioned, if the game is a draw but one player has a knight advantage, its clear that the majority of cases, this points to greater strength (not many outlier cases, but in general). If I have taken every piece of my opponent but then stalemated him, again the additional material points to greater strength, again not in all cases. My point is the end result a drawn the game provides little information, however there is much meta data in the moves of the game, for instance, who is in control (or calling the shots), who is steadily gaining a better position, which player is keeping the initiative throughout the game. Again in a lose/win scenario there is much meta data to be derived, one example is game length (which also is valuable information forof the drawn games). If a player loses in less moves, the gap of strength is greater than losing in less moves, however with greater strength players the minimum moves in a lost game increases. Someday the chess world will wakeup and realize the plethora of meta information available is statistically telling enough to make generally accurate predictions off of a few games, 1 game and even a sequence of moves. Until a paper is released proving this obvious logic, haters gonna hate.

[@JoeKempsey]

@@Zatchurz Very interesting concept! Even though the game ended in a seemingly typical draw, there is so much insight able to be taken away from each and every move

[@basantsharma4857]

Wow nice video crazy how many thread Stockfish 16 NNUE

[@JoeKempsey]

5 threads for this video! Cheers and thank you for watching :D

[@bhgtree]

What did Leela say when she heard that SF 16 came out?? "Another fish to fry."😂

[@JoeKempsey]

😆😆 Stockfish 16 vs Leela v29 would make for quite the interesting encounter!

[@metroexodus4388]

Thank u 🎉🎉🎉

[@JoeKempsey]

Thank you for watching! :D

[@metroexodus4388]

@@JoeKempsey 🤩🤩

[@metroexodus4388]

@@JoeKempsey I ll text u in ur email.

[@BusyBee-n4s]

Hey could you nerf stockfish 15 to play openings like french defence and mordern defence against sf16?

[@JoeKempsey]

Thank you for your suggestion! French defense would be intriguing, especially considering how earlier SF versions used to struggle badly with it

[@tenningale]

Think SF16 would probably beat SF15's 1...g6 a big percentage of the time but would love to see that type of game. 1...e6 should hold up better but would also be fascinating to see both run at a high sample.

[@Engine448]

hey you should probably not use lucaschess because i heard that they limit the engines strength.

[@JoeKempsey]

Hmm interesting, never would think it would but who knows

[@Engine448]

@@JoeKempsey yea when i used lucaschess stockfish averaged 95 accuracy with tons of thinking time and then i switched to banksia gui and i average 98-100

[@JoeKempsey]

@@Engine448 Ooh

[@basantsharma4857]

@Engine448 and @Joekempsey Lucas chess the stockfish 16 is real hash size 128 and thread 5 defeat checkmate move 49 Stockfish 15 Best Stockfish 16! Hash size 128 and thread 5

[@Engine448]

@@basantsharma4857 ?

[@Sack_Zement]

Could you please include the hardware specs in the Infobox?

[@JoeKempsey]

Hi! I’m using an Acer Aspire 5 (8GB ram, 512GB SSD), with an AMD Ryzen 7 3700U dual-core processor. Will add to description. Cheers!

[@JoeKempsey]

Also to note, I have +8GB RAM stick, so 16GB total RAM

[@김시현-n7o]

Best fish ever

[@JoeKempsey]

Yeppp :)

[@jaysinghthakur3519]

How to download stock fish 15& 16 plz tell somebody

[@JoeKempsey]

Hi, check out www.stockfishchess.org to download stockfish :D Then you just need a GUI (graphical user interface) to run stockfish on. Check out link below in case need help on setting this up! m.kzbin.info/www/bejne/d6STk5KwnbCibpo

[@intel7772]

Hi!Can you play stockfish 16 vs Magnus Carlsen with knight odd

[@JoeKempsey]

Hi! We shall see :) I wish the Play Magnus bots were on chess.com only since it’s tricky to set up on the app. But will at least do some testing of SF16 against the Play Magnus bots!

[@happygood18]

why 256 mb hash size?😂😂

[@JoeKempsey]

😅 With such little per move Stockfish doesn’t need a large hash size, so even using a super low hash is okay

[@happygood18]

@@JoeKempsey 🤨🤨🤨✂️

[@seblubla1685]

Why did you let it play as white?

[@j.21]

to give white better chances to win

[@JoeKempsey]

Hi, this was done to get the best chance possible at a decisive result - and to put Stockfish 16 in the spotlight Will certainly do a rematch with colors reversed if interested! Cheers :)

[@j.21]

my "a" comment 2 hours ago disappeared?

[@JoeKempsey]

Not sure :( I never delete comments so must’ve been auto-removed by KZbin

[@examiningdata7952]

bro , do a face reveal !

[@JoeKempsey]

Older face reveal in this video :) kzbin.info/www/bejne/iKvQfp-qoJ6paMU I could do some videos with commentary and face cam honestly, just would have to get it set up