Kramnik has a point

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

Uri Blass
Posts: 10420
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Kramnik has a point

Post by Uri Blass »

BrendanJNorman wrote: Tue May 07, 2024 12:13 pm
towforce wrote: Tue May 07, 2024 11:45 am
BrendanJNorman wrote: Tue May 07, 2024 10:47 am
towforce wrote: Sat May 04, 2024 12:34 pm The pre-moves are impressive, but so is something else: look how relaxed Hikaru looks! I'm nowhere near that relaxed when I'm playing at "game in 10 minutes" time.
Not that impressive imo.

Nakamura has only *ever* played chess and was already playing "kindergarten championships" at 5 years old, coached by his FM stepfather.

He has now had decades of coaching, training, tournament experience, computer analysis and one of the best learning environments on the planet (he was already playing in the Marshall Chess Club at age 10).

Chess is similar to languages in that if you gain fluency at a younger age, you will not have an "accent" - you will flow in and out of this language like a native speaker.

This is Naka's current chess situation.

31 years of dedication (he's now 36 years old) while still being relatively young - this makes him look very impressive to those who have only put in a portion of the time and effort.



Reminds me of the piano maestro who was once having her play admired by an onlooker.

When she finished the onlooker said "OMG I would do ANYTHING to play like you do".

To which the maestro responded "No you wouldn't."

The maestro was right.

Very good point.

I remember when Karpov would play in world championships (classical time control), he would lose a lot of weight, and this was likely due to stress. You seem to be telling me that high level chess at very short time controls is basically reflexive response. :)
Yup. High level human chess at fast time controls is 99% intuition built up by learning thousands of chess patterns over years of study.

The last 1% is pausing for a sec as a position becomes tactical to quickly check/calculate whether something "works", or to calculate a response when faced with an unpleasant surprise from your opponent.

In CC terms, humans are 99% NN eval with a very slow search and tons of pruning. :lol:
I disagree about it.
I do not think engines do less pruning than humans.
Engines also do tons of pruning but some of it is clearly a dubious pruning relative to humans.

[d]2r2rk1/pq3ppp/3bpn2/1p6/2PNP1PP/P3BP2/1PQ5/R3K2R w KQ - 0 18

I expect even humans with rating 2000 at blitz to see that Nxb5 is a bad capture because Qxb5 is a good capture
Stockfish NNUE does not understand it and at small number of nodes it suggest Nxb5 because it prune Qxb5.
BrendanJNorman
Posts: 2559
Joined: Mon Feb 08, 2016 12:43 am
Full name: Brendan J Norman

Re: Kramnik has a point

Post by BrendanJNorman »

Uri Blass wrote: Thu May 09, 2024 12:35 pm
BrendanJNorman wrote: Tue May 07, 2024 12:13 pm
towforce wrote: Tue May 07, 2024 11:45 am
BrendanJNorman wrote: Tue May 07, 2024 10:47 am
towforce wrote: Sat May 04, 2024 12:34 pm The pre-moves are impressive, but so is something else: look how relaxed Hikaru looks! I'm nowhere near that relaxed when I'm playing at "game in 10 minutes" time.
Not that impressive imo.

Nakamura has only *ever* played chess and was already playing "kindergarten championships" at 5 years old, coached by his FM stepfather.

He has now had decades of coaching, training, tournament experience, computer analysis and one of the best learning environments on the planet (he was already playing in the Marshall Chess Club at age 10).

Chess is similar to languages in that if you gain fluency at a younger age, you will not have an "accent" - you will flow in and out of this language like a native speaker.

This is Naka's current chess situation.

31 years of dedication (he's now 36 years old) while still being relatively young - this makes him look very impressive to those who have only put in a portion of the time and effort.



Reminds me of the piano maestro who was once having her play admired by an onlooker.

When she finished the onlooker said "OMG I would do ANYTHING to play like you do".

To which the maestro responded "No you wouldn't."

The maestro was right.

Very good point.

I remember when Karpov would play in world championships (classical time control), he would lose a lot of weight, and this was likely due to stress. You seem to be telling me that high level chess at very short time controls is basically reflexive response. :)
Yup. High level human chess at fast time controls is 99% intuition built up by learning thousands of chess patterns over years of study.

The last 1% is pausing for a sec as a position becomes tactical to quickly check/calculate whether something "works", or to calculate a response when faced with an unpleasant surprise from your opponent.

In CC terms, humans are 99% NN eval with a very slow search and tons of pruning. :lol:
I disagree about it.
I do not think engines do less pruning than humans.
Engines also do tons of pruning but some of it is clearly a dubious pruning relative to humans.

[d]2r2rk1/pq3ppp/3bpn2/1p6/2PNP1PP/P3BP2/1PQ5/R3K2R w KQ - 0 18

I expect even humans with rating 2000 at blitz to see that Nxb5 is a bad capture because Qxb5 is a good capture
Stockfish NNUE does not understand it and at small number of nodes it suggest Nxb5 because it prune Qxb5.
Feel free to disagree, but...

From my perspective you are ignoring the huge gap between human and engine play.

You use the phrase "even humans with 2000 rating" as if this is low.

It is (very) low for computers sure, but 500 Elo points (!) above the average human blitz rating of 1500.

(source: https://lichess.org/stat/rating/distribution/blitz)

The average human chessplayer of 1500 is actually *very* likely to play Nxb5, missing Qxb5).

I have seen players even higher missing even more obvious things at blitz.

And LTC.

We all remember Kramnik missing the simple mate in 1 vs Fritz, and Ivanchuk also missed a mate in 1 vs Anand once.

This week Carlsen missed a simple mate in 1 vs Nakamura.

Would an engine miss a mate in 1?

Never.

Because they *still* have enough search to see something so simple, but human beings are somehow able to prune so much in a position, that they somehow subconsciously judge the mate threat (or whatever) as irrelevant.

Until it happens and it's over.

Isnt this by definition what pruning is? (I'm not a programmer, so not 100% sure)

Human pruning seems far more extreme to me.
User avatar
towforce
Posts: 11706
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK

Re: Kramnik has a point

Post by towforce »

Peter Berger wrote: Thu May 09, 2024 9:52 amBtw: the problem with Hikaru's streaks is not that they exist, but their frequency.

1. Chess.com have said that they've had the case examined by a highly qualified statistician (linked from earlier in this thread)

2. The data being used by Kramnik is obviously cherry picked. This is incompatible with drawing inferences from data using statistical distributions

3. Statistical distributions require the events to be independent. Please explain to me how the same two players playing each other 50 times in a row at a game of skill represents 50 independent events. I can't wait to read this! :)
The simple reveals itself after the complex has been exhausted.
User avatar
towforce
Posts: 11706
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK

Re: Kramnik has a point

Post by towforce »

towforce wrote: Thu May 09, 2024 3:47 pm
Peter Berger wrote: Thu May 09, 2024 9:52 amBtw: the problem with Hikaru's streaks is not that they exist, but their frequency.

1. Chess.com have said that they've had the case examined by a highly qualified statistician (linked from earlier in this thread)

2. The data being used by Kramnik is obviously cherry picked. This is incompatible with drawing inferences from data using statistical distributions

3. Statistical distributions require the events to be independent. Please explain to me how the same two players playing each other 50 times in a row at a game of skill represents 50 independent events. I can't wait to read this! :)

A question: it's easy to come up with ways that Niemann could have cheated in OTB games (not saying he did - just saying that it's easy to see how it could have been done), but in the case of these high speed games, how would it even be possible to cheat?

One thing I can say with confidence: if he's hacked the Chess.com UI, he didn't do it himself: he's not a computer hacker. Whomever did the hack did a damn good job getting all those pre-moves to work so well.

Also, in these streaks, is there an absence of human error (the kind of mistakes people make but computers don't)? That would be good evidence.
The simple reveals itself after the complex has been exhausted.
BeyondCritics
Posts: 406
Joined: Sat May 05, 2012 2:48 pm
Full name: Oliver Roese

Re: Kramnik has a point

Post by BeyondCritics »

Peter Berger wrote: Thu May 09, 2024 9:52 am Btw: the problem with Hikaru's streaks is not that they exist, but their frequency.
You can sell this as preliminary data screening and i think this might be credible, but this has to be done absolutely silent. We are talking about one of the best blitz players ever and are bound to expect the unexpected.
Presenting calculated probabilites for home grown events after the fact is just nonsense and a mere proof of incompetence. Because of this alone, i would trust nothing from team Kramnik.
Since if you are multiplying are large number of arbitrary probabilites together you are always bound to get something that approaches zero exponentially fast.
I have also seen studied data scientists trying their hand on the Nieman case, producing bullshit also.
In this matter you should really trust only someone wo has studied mathematics and statistics.
Chess.com explained they did simulations on this and found nothing. Makes sense to me.
CornfedForever
Posts: 634
Joined: Mon Jun 20, 2022 4:08 am
Full name: Brian D. Smith

Re: Kramnik has a point

Post by CornfedForever »

BeyondCritics wrote: Thu May 09, 2024 8:04 pm We are talking about one of the best blitz players ever and are bound to expect the unexpected.
Nakamura is similar to that proverbial monkey who sits at a keyboard, day after day, week after week, month....(well, you get the idea) typing away -- only instead of coming up with the improbably word...sentence...Shakespeare play, Nakamura comes up with 'improbable' streaks. Add to that, Nakamura (gets paid for it!) and is capable of learning and applying (and choosing whom to avoid) unlike said illiterate monkey.

The improbable simply becomes reality.

Incidentally, there has been quite an outcry against Kramnik's latest accusations against Nihal Sarin - "mathematically 99.9999% impossible".

As Caruana said of Kramnik not long ago, " He's looking at the chance that someone will achieve a result in a given series of games and conflating this with the chance that someone is cheating. But these are separate issues".
Peter Berger
Posts: 663
Joined: Thu Mar 09, 2006 2:56 pm

Re: Kramnik has a point

Post by Peter Berger »

towforce wrote: Thu May 09, 2024 4:38 pmA question: it's easy to come up with ways that Niemann could have cheated in OTB games (not saying he did - just saying that it's easy to see how it could have been done), but in the case of these high speed games, how would it even be possible to cheat?
Your question is a bit off-topic, and it took me some time to realize, that I can easily give a kind of relevant and true answer.
I operated Crafty in the Blitz event in Ramat-Gan in 2004. The time control on the physical board was 5 0. Crafty was running in text mode, its internal clock was set to 3 0. There were a few very long games, I could do that, I was faster than the GUI guys.
Only a year later I gave this setup another try and I couldn't understand how I could ever have done this, it felt completely insane and not doeable at all.;)
I don't want to discuss potential cheating methods, this feels wrong to me. But here the same would apply: that you can't imagine how you could do this doesn't mean it can't be done.
User avatar
towforce
Posts: 11706
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK

Re: Kramnik has a point

Post by towforce »

Peter Berger wrote: Fri May 10, 2024 8:30 pm...that you can't imagine how you could do this doesn't mean it can't be done.

You seem to have a grim persistence in your feeling that Hikaru's streaks are unnatural, so why don't you do something that Kramnik should have done, and look at ALL of the evidence?

My suggestion:

1. get a tool that enables you to easily analyse multiple games (Chessbase? See thread on analysis tools for other suggestions)

2. load up some games he lost, and some games that he won in a streak

3. See how many mistakes he makes in games that he loses, and how many mistakes he makes in games that he wins in streaks

The point is, if he usually makes multiple mistakes in games he loses, then you'd expect him to make mistakes in games he wins as well.
The simple reveals itself after the complex has been exhausted.
Alexander Schmidt
Posts: 1217
Joined: Thu May 10, 2007 2:49 pm

Re: Kramnik has a point

Post by Alexander Schmidt »

Peter Berger wrote: Fri May 10, 2024 8:30 pm But here the same would apply: that you can't imagine how you could do this doesn't mean it can't be done.
If you cheat in Blitzgames, you have to concentrate on whatever you helps cheating. You cannot explain all of your moves and make premoves while concentrating on cheating.

Beside that chess.com evaluates the movetimes. You will get caught quite fast if you just transfer engine moves to a game on chess.com because your movetimes will indicate this.

Beside that chess.com evaluates the moves and compares them to enginemoves. If you avoid this and sprinkle in own moves (while not 100% concentarting on the game) you will make blunders.

If you avoid all this cheating might help a little bit on top level, but will not help you make huge winning streaks.
User avatar
towforce
Posts: 11706
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK

Re: Kramnik has a point

Post by towforce »

Alexander Schmidt wrote: Sat May 11, 2024 8:06 am
Peter Berger wrote: Fri May 10, 2024 8:30 pm But here the same would apply: that you can't imagine how you could do this doesn't mean it can't be done.
If you cheat in Blitzgames, you have to concentrate on whatever you helps cheating. You cannot explain all of your moves and make premoves while concentrating on cheating.

Beside that chess.com evaluates the movetimes. You will get caught quite fast if you just transfer engine moves to a game on chess.com because your movetimes will indicate this.

Beside that chess.com evaluates the moves and compares them to enginemoves. If you avoid this and sprinkle in own moves (while not 100% concentarting on the game) you will make blunders.

If you avoid all this cheating might help a little bit on top level, but will not help you make huge winning streaks.

Suppose circumstances arose which forced you to help someone cheat in online blitz games: the only way I can think of to do that would be to take control of the Chess.com game playing UI, and link it to a chess engine which would choose the moves and relay them to Chess.com (either through Chess.com's UI or through their web services using a fake Chess.com UI for the YouTube viewers).

If this were done, all of the moves would be chosen by the engine: in this scenario, you'd want the engine to also make some human-like blunders to fox anyone analysing the games, but not any blunders which are bad enough to allow the opponent to win.

A bit of a software development project - but possible.

Which is more likely - that the above software was built and that Hikaru is using it, or that Kramnik's analysis is flawed? For me, an easy question: I have already worked out that Kramnik's analysis is flawed, and this has been confirmed by an expert statistician.
The simple reveals itself after the complex has been exhausted.