The Legendary 270 Move Game! Rybka vs Nakamura Game Analysis 

Computer: Noooooo!!!! You can't keep shufling without doing anything. Naka: haha, so i went la la la la la laaa

I like how he just bullies the computer at the end

mating with 5 bishops is required technique for any grand master

"2008 so it wasn't actually that long ago" that was 11 years ago lol

I can't attack a computer because he'll smoke me like a clown LMFAO

The "not attacking chess engines" advice was useful to me because everytime i found a great aggressive attacking move against an engine it would just calculate perfectly and stop it in an instant.

Naka is a wizard, he can turn a horse into a golden goose.

Hikaru: "6 bishops to mate with" Me: Huh.......

Just to clear up some stuff about AI: The computer is not necessarily looking a set amount of moves ahead, it's probably doing some sort of iterative deepening in its game tree, however, how far ahead it can calculate is dependent upon how well it can prune out bad paths. The pruning potential is limited by how well it can rate a given board state, so the trick is to get it into a board state where its heuristic evaluation is unreliable, hence the rooks vs bishop/knight situation. Because the heuristic was most likely designed to value those pieces more, it results in a suboptimal evaluation of those board states, and by extension, weak game tree pruning and it ends up having to calculate more paths than it needs to and can't reach as deep a depth. Chess AIs strive off positions with lots of trades and captures where they can very efficiently prune out bad paths because their heuristics are designed for exactly that, not super awkward locked board states where they're up material but can't actually do anything with the advantage. In Naka's game, it probably never bothered calculating the lines where he gave up his rooks because it never expected him to go down those paths because according to its heuristic Naka was way worse and had no reason to make those moves. Neural networks like Alpha Zero still use game trees, pruning, and all the other stuff that comes with it, but their key strength is derived from the fact that they don't use human designed heuristics. Instead, programs like Alpha develop their own heuristic by playing themselves millions of times over, and therefore they aren't limited to the quality of human designed heuristics that don't work as well in these unexpected board states. Alpha almost certainly wouldn't make the same mistakes this AI did because it wouldn't blindly evaluate rooks as better, despite that being true in 99% of games. Nobody hardcoded into Alpha that rooks are better than bishops and knights, it learned it from trial and error, and as a result it also learned when they're not better, so it makes far better moves in these unexpected positions. Neural networks are very much reminiscent of how the human brain works, and thereby don't fall victim to the kinds of miscalculations these other engines do. To explain how I would interpret the terms used: The contempt factor is the AI being mislead by bad board state evaluations like I mentioned before with the rooks, since the heuristic designates them as arbitrarily higher. The horizon effect is the inability to reach a reasonably deep search depth brought on in part by the contempt factor. The bad board evaluations make it really hard to sort which moves to look at first and prune bad lines in an efficient manner, so it has trouble reaching a reasonable depth. The pacman effect is not something I'm familiar with, but I'm certain an AI for pacman would be completely different from chess engines like these. It's probably just used as another term for the previously mentioned concepts. Just thought I'd write this up because I see some people trying to describe what's going on here while missing some of the fundamental concepts of these kinds of AI.

Imaging being skilled enough to troll rybka😁

The greatest game hikaru ever played was the "lalalalalaallalalalala" game

I finally got it!! The only thing missing from my games is "la la la la la" Once I incorporate that I'll be as good as Naka

4:47 Those King moves were so careless that is actually hilarious xD

7:36 when the physics teacher starts to explain something in class

lol. "What was that!?" he says trying to undo the spoiler he just leaked. 9:05

I remember this game. Nakamura also did a hilarious game with 6 knights checkmate

He makes frequent references to "contempt factor", "pacman effect", and "horizon effect" but doesn't actually explain what they are. But I do know they "came into play" :)

i stalemated my opponent just by looking at those bishops

Imagine playing a grandmaster and cheating using an engine, and they still win

Computer: Im up material Hikaru and his bishops:

Me: Trying to re-maneuver pieces in 15 moves Hikaru: that's not how it works, just sing along with me- lalalalala...

I could actually see this contempt bug being a cool plot point in science fiction.

Rybka became the best engine in 2005. When this game was played, Rybka was still the best engine. The era of Houdini hadn't begun yet.

9:04 - 9:20 “Sorry about that” -Antonio Nakamura

So this game was taught us about chess influence, there was pac-man effect and horizon effect, but also agadmator effect at 9:18

Still theory.

Hikaru: *casually headbanging to LALALALA while exploiting bugs*

2025: AI takes over the world HUMAN: secret wooden shield techniques

If I understand well: Horizon effect : a close position that a human will take advantage of in the long term (impossible to calculate but a human can see that the opponent pieces will be stuck while his will be able to make an attack. Pac-man effect: the computer is up material in a drawish position so it does a breakthrough except this allows the human to have a winning endgame due to the pawn structure and the piece disposition. I believe none of these would affect a good chess neural network (and that is what deepmind claims to have with A0)

Great to have these excellent videos to watch from a Super Grand Master.

AI trying to take over the world Hooman: lalalalalalala *(checkmates casually)*

8:00 imagine if he did one more bracket

I am doing lalalalala but still lost everytime vs Rybka lol

I remember this game in 2008 and the computer evaluates the position like +1.5 or something and then pushes c4 to avoid a draw. They would usually set a contempt factor against a draw to 0.5 or 0.0 so if the computer had a 1.5 it would never allow a draw (by 50 move rule) and would make the only other legal move - c4.

Rybka mean fish D:

aaaah come on Nakamura, your "most LEGENDARY" game was against Crafty Computer where you were "horsing around" with six knights)))

He mentions the Pac-Man Effect several times but doesn't explain it and google turns up nothing.

This is the funniest showcase of a fortress I think I’ll ever see

9:50 "So, six bishops to mate with" - Damn! Didn't expect those kinds of dirty orgy ideas from Hikaru... Too much man...

How to win stockfish: lalalalalala

"wasnt actually that long ago" > more than 10 years ago

Imagine this man was about to beat the strongest engine at the time and he accidentally stalemates with 5 bishops on the board

Agadmator: capture capture capture Hikaru: shuffle shuffle shuffle Lol

"Six bishops to mate with" -Nakamura 2019

10:30 tap start to skip cutscene.

He BM'd the computer to assert human dominance.

Your my favorite chess player. I play exactly like you... Except i lose every game.....

From the first time I saw this game I was instantly impressed! You played a brilliant game !!!

That was clever of you but you are confused about the meaning of "contempt" in computer chess. If computer has contempt=0.25 pawn (say) that means it regards the value of a draw NOT as zero, but rather as -0.25. Therefore it will not take a draw unless it thinks it is behind by at least 0.25 pawns. In the present game, Rybka figured it was up two exchanges, +4, so therefore it was not going to take a draw and would sac up to about 3 material to avoid a draw. This had nothing to do with "contempt" being nonzero; it was simply that its evaluation function is material-dominated, and its inferior long-term understanding of this kind of endgame compared to Nakamura's understanding. And this is not an easy "bug" for Rybka's authors to repair, because the only cure is to make it understand chess better, which is not a quick easy fix. It also was key that long term understanding was needed, i.e. beyond its search horizon. Nakamura was looking 50 moves ahead, in a sense, i.e. "in 50 moves, the computer will think it needs to sac a pawn to avoid a 50-draw, and I will predict which pawn that will be." And he understood the power of his pawns were more than material. The computer could see only maybe 5-10 moves ahead, and its eval was too material-dominated, so it was blind to Nakamura's whole plan until too late. Contempt is still a necessary ingredient of most chess programs today. Alphazero however has a very good understanding of chess in is neural-net evaluator, quite likely better than Nakamura's understanding of chess. So I think he's got no chance versus Alphazero. Versus an excellent handmade evaluator like Stockfish, Nakamura might be able to think of some way to exploit stockfish's inadequate understanding of something, but it gets harder because stockfish's programmers gradually make it understand more chess better, so there is less and less that can be exploited. But if you look at games where Lczero and Alphazero are killing Stockfish, you will see a fair number of games where they did it in a pretty "human" way, i.e. they understood something, which is human-understandable, which Stockfish's evaluator just did not evaluate correctly. In particular, long-term imprisonment of pieces due to a combination of blockages and being tied down to duties - Alphazero could understand that much better than Stockfish. And "thorn pawns"; Alphazero understood their power much better. And "fortresses" are another thing Stockfish understands poorly. Also, Stockfish did not understand ...c4 in the French , but most or all human grandmasters do.

last moments in human > machine history

A subtle bongcloud at 4:47

I want agadmator to make a video on this.

Was this the last time a GM was able to beat one of the top programs at the highest level?

Hikaru is clogging the toilet

He bullied a computer😂😂

7:08 actually the computer played c4 in order to avoid draw (50 move rule)

I guess that’s what we get from a super GM. explains everything extremely quick & it’s hard to keep up

I love Naka's videos, but it's kinda annoying that he get's constantly interrupted by the chat

The flexing is real.

Wtf this man is finding bugs and making the devs do patches for a thousand year old game...I mean...wuttttt

Very nice of you Naka, your psychological trick worked! I've learned a lot from you

Humanities last stand and win vs the computer

I would not want that a pgn

greatest weapon in chess is la la la la la la la

no one: hikaru: so i went lalala laa bababa baaa

Ur the only RU-vidr that clickbaits and then admits to it 5 seconds into the video lmao

Legendary... I think most players wouldn't think to lock up the game in such a way hikaru did and it was two rooks vs a bishop and a knight. His level of understanding for thw fame is truly incredible. Im still waiting for the day you turn number 1 Naka, maybe if im lucky ill get ti lay you some day...

shuffling pieces around in a closed position drives ribka crazy! 😁

Just trying to follow Naka's fast analysis...

Old Hikaru videos have a strange relaxing effect. Currently, his speech pattern is similar to being hit on the head with a hammer in comparsion.

gelfand vs nakamura analysis needed..this young nakamura player is pretty nice

I was just scrolling some comments and looked up randomly so see *THIS* 7:43 Me: woah woah woah wtf is going on?!

Is this the last game ever when Human (if Naka is :) win against top program ?

My longest game in life , I mean the game which I played the most moves , I played a game with 240 moves !!!!!!!!!!! The Longest game I have ever played in my life till now !!!!!!!!!!!!!!

I knew it ! I have always said the human (GM Hikaru) smarter than machine

another level of dope

random jump to the 5 bishop mate lmao

“TheLegend27”

Love this guyyyy

Apparently 2008 wasn’t that long ago.

I seem to recall reviewing this game after it happened and determined the operator must have set the contempt to avoid drawing at all costs. When I ran it through default settings Rybka at the time, it KNEW the draw was coming and chose to accept it. So in essence, this wasn't a true exploit of standard computer contempt, but rather of operator changing the parameters to avoid the 50 move draw rule. For example the blundering pawn break on c4 it evaluated as losing on the spot, so quite obviously it knew it was a losing move.

Whatever... he beat a supercomputer in 2008 bc he outsmarted it.

BOUM G4 !!!

Hikaru smoked Rybka like a clown

more cowbell

LEGENDARY!

It's time for Leela! I just kidding :D

Rybka in Polish is Fish like stockfish 😉

Rybka means fish :O

I searched it because i found it's reference on a book named Algorithms to live by

I still have no clue what the pacman effect is supposed to be and Google isn't helping much.

I don't even understand what the hell he was talking about. All I see is he bullied the computer.

i hope he analize the one with the six knights

"The chances become higher, much more astronomical." Loved the analysis, but did Hikaru learn the word "astronomical" the day before this video and then wanted to show off his knowledge?

End is hilarious!!! :D

Whats rybka's elo 3000?

Shuffle it up lol

how do you remember all of the moves 11 years later?

HIKARU is awesome

Still dont get it, whats the difference/relations between contempt, horizon and pacman effect??