Authors of chess engines getting crushed by their creations

[gonzochess75]

If you wanna see the chess engine authors embarrassed, then check out this article by the_real_greco

https://www.chess.com/blog/the_real_gre ... ine-part-i

[xr_a_y]

Of course, the same holds for me and Minic (or Weini ...)

https://lichess.org/GFIk9Tm4/black

[bob]

Not sure this is "news". In the early 80's, I might win one out of 100 games against Cray Blitz, yet our ratings were pretty equal in terms of USCF. 5 years later, that was zero out of 100. Been that way all the way through current Crafty. It is really easy to develop an engine that beats you trivially. I remember an old Star Trek episode where Spock proved the computer in the Enterprise had been tampered with because he beat the chess program he supposedly wrote, in several successive games. He stated "the best I should be able to achieve is a draw." We now know that kind of "logic" was badly flawed. And still is.

Looks like hardware improvements turned the old simplistic search approach into a super-GM, and it now seems that the ANN programs are now beginning to reap the same benefits in terms of massively parallel GPU cards, so that the old issues that killed them (speed related and training data sizes) is now a non-issue.

I'm not convinced ANN programs will ever really surpass the traditional alpha/beta algorithm, with any significant margin, but it might end up being just as good...

[silentshark]

I agree, Bob, this is not news. Just checked, and 20 years ago, in 1999, I took my program to the local chess club to play.

At this point, it was running on a Pentium 133Mhz processor, which is many times slower than processors today.

Anyway, back then it beat a few players rated between 1700 and 2100, and was comfortably better than me (circa 1500).

I was delighted to see the 'trip down to the chess club' is online - http://www.chesscomputeruk.com/SS_81.pdf

[Leo]

Not sure this is "news". In the early 80's, I might win one out of 100 games against Cray Blitz, yet our ratings were pretty equal in terms of USCF. 5 years later, that was zero out of 100. Been that way all the way through current Crafty. It is really easy to develop an engine that beats you trivially. I remember an old Star Trek episode where Spock proved the computer in the Enterprise had been tampered with because he beat the chess program he supposedly wrote, in several successive games. He stated "the best I should be able to achieve is a draw." We now know that kind of "logic" was badly flawed. And still is.

Looks like hardware improvements turned the old simplistic search approach into a super-GM, and it now seems that the ANN programs are now beginning to reap the same benefits in terms of massively parallel GPU cards, so that the old issues that killed them (speed related and training data sizes) is now a non-issue.

I'm not convinced ANN programs will ever really surpass the traditional alpha/beta algorithm, with any significant margin, but it might end up being just as good...

Welcome back Bob.

[gonzochess75]

It might not be news, but I thought it was fun and others would like to see some engine devs (like me) getting embarrassed. BTW, so cool to have Bob commenting. I'm a big fan

[adams161]

Not sure this is "news". In the early 80's, I might win one out of 100 games against Cray Blitz, yet our ratings were pretty equal in terms of USCF. 5 years later, that was zero out of 100. Been that way all the way through current Crafty. It is really easy to develop an engine that beats you trivially.

I think i can still maybe draw my Pulsar engine at it's hardest level. It only got up to about 2600 ICC Blitz. But for analysis of users Pulsar games in my App Pulsar Chess Engine for iOS, i'm using Crafty which i think is over 500 rating points stronger or more on computer list than Pulsar. And in my view while not Stockfish and i'm even a few versions behind modern Crafty, it's strong GM strength per the rating for the version i'm using, about 2700 computer rating I found when i Googled the version. Current at the time was nearly 3000.

They can actually play it and Pulsar too even though Crafty is only available for Analysis. Since summer 2018 when i had Crafty then implemented only in openingtree, there has been a "Move now" button. And I designed ti to simulate play if they want to play it out. It replaced analysis with something like "Move made: e4", and stops the engine analysis, which they can turn back on, sort of inviting people to play if they want. And some have written me about their Crafty games in OpeningTree. And now it's in Pulsar iOS as of this last Friday, https://itunes.apple.com/us/app/pulsar- ... 40447?mt=8

Did my first book building at the time in 1998, just a few months before i started Pulsar, and the program i used was crafty making books from specific pgn files of only a few openings. And back in 1998 using a then older version of Crafty, on a pentium 133, I did beat Crafty once. I showed my then chess coach and he counted how many moves ahead Crafty had to think to stop the attack and commented it looked like i got just above the horizon. Only time i beat it. I doubt i can get as lucky again

Edit: I should add that I asked for permission to use Crafty in my apps on December 3, 2014, and it's not automatic the right to use Crafty just because it's open source. And conditions i find reasonable things like name your engine to your users and don't enter more than one copy in a tournament etc.

[chrisw]

Creation is too strong a word. The creative person is the one who first did something. The later ones follow a recipe.

[adams161]

Creation is too strong a word. The creative person is the one who first did something. The later ones follow a recipe.

Bob Hyatt, the Crafty author, was a pioneer of parallel computing.

Myself when i started Pulsar didn't have much claim to fame i was not following a recipes. But I wanted to do something different so I made a game plex. Pulsar for iOS and Android plays Chess, Atomic, Crazyhouse, Three Checks, Loser's Chess and Giveaway. And i spent time with all of them working on how to teach the chess code to play variants. I did start with chess code some parts stripped out.

So i think there is different types of creation even if following an area others are working in.

[chrisw]

Creation is too strong a word. The creative person is the one who first did something. The later ones follow a recipe.

Bob Hyatt, the Crafty author, was a pioneer of parallel computing.

Myself when i started Pulsar didn't have much claim to fame i was not following a recipes. But I wanted to do something different so I made a game plex. Pulsar for iOS and Android plays Chess, Atomic, Crazyhouse, Three Checks, Loser's Chess and Giveaway. And i spent time with all of them working on how to teach the chess code to play variants. I did start with chess code some parts stripped out.

So i think there is different types of creation even if following an area others are working in.

I wish to make a distinction between going down an unproven path that nobody has done before, and doing something that has been done before. The former is uncertain and risky and coming out the other side with something functional is the creative achievement and the creative process.
Going down a path that has already been proven, whether or not the problem solving involved was redone or looked up, is an entirely different affair.
It takes a very special personality and determination to go where nobody else has been, and we need/use a special word to describe it. That word is creativity. It doesn’t apply to something that’s already been knowingly done elsewhere.