Fwd: Open letter to the CSVN

[M ANSARI]

There is a huge difference when you claim that an engine participated in a tournament and when it plays a few games due to a technical failure. Not portraying the entire episode is equivalent to using snippets of speech out of context to make up your own story. There is no need to morph the truth, just say it as it is. With respect to what constitutes "plagiarism" as per ruling 2 of ICGA ... there clearly needs to be a defined method of what constitutes plagiarism. If using knowledge screened from open source engines is not OK, then it should be quantified in a manner that is easy to assess. Otherwise you will get into the mess computer chess is at the moment.

[Graham Banks]

It's pointless to argue with closed-minded people.

To be fair, this could be exactly how those who disagree with the ruling could feel too.
And no, I'm not taking sides, just pointing this out.
Too much nastiness from some on both sides.

[Evert]

how do you explain the 2.200 ELO jumpings from so many amateur programmers.

It is not very hard to make a chess program that plays at 2200 elo (on CCRL scale). A good search with a reasonable but basic evaluation will get you there fairly quickly.

[hgm]

Indeed, a 2200 Elo jump from 0 to 2200 is not at all the same as one from 1000 to 3200. It is a bit like comparing athletes on the 100m dash. I could run that in 30 sec, but after a month of training I cut that down to 20 sec. That is 10 sec off. So if someone who always used to run it in 15 sec a month later suddenly runs it in 5 sec, there is no reason to suspect he is using bionic implants, right? He just got 10 sec of his time by training, like I did...

[Frank Quisinsky]

Hi there,

thinking on the first Winboard engines, the first rating list we create. Around 8-9 years before CCRL. Not easy to reach 2.200 - 2.300 ELO was the answer from so many amateurs.

Some of the amateurs wrote that 2.500 ELO is not to beat. We had in this times Crafty, Zarkov and Nimzo, Shredder 3 which are winboard compatible with 2.500. Today many started with 2.300 - 2.500 ... clones or open ideas!

Nice example:
Today no problem
Or are the programmers today more intelligence as in the time we powered Winboard at first?

Best
Frank

[michiguel]

how do you explain the 900 elo jump between rybka versions.

do you know any chess program (that was developed by programmers not by copy-paste enthusiasts) that ever made this kind of strength increase ?

Ruffian

Miguel

[hgm]

Some of the amateurs wrote that 2.500 ELO is not to beat. We had in this times Crafty, Zarkov and Nimzo, Shredder 3 which are winboard compatible with 2.500. Today many started with 2.300 - 2.500 ... clones or open ideas!

Nice example:
Today no problem
Or are the programmers today more intelligence as in the time we powered Winboard at first?

I don't think this is a mystery at all. Before the CCRL existed there obviously was noCCRL rating scale, so the ratings you quote must heve been from another scale, which makes it doubtful that they can be compared to current CCRL ratings. If the 'unbeatable 2500' was relating to a Human GM rating of 2500, hardware is not factored out. On 20-year-old hardware Houdini would of course also not be above 2500 Elo.

And then there are of course the 'open ideas' that every one uses, and which give a huge Elo boost to simple programs: alpha-beta, QS, MVV/LVA move sorting, null move, LMR, hash table, killer heuristic, mobility eval. Implement all that correctly, and you are already at 2500 Elo CCRL. Micro-Max, with only 100 lines of code, only implements the most necessary part of that (no real move sorting, no killer, no mobility, no passer evaluation, no piece-square tables), and it is already around 2050 Elo CCRL.

These open ideas are not tied to any particular Chess engine,andalmost everyoneuses them; you can learn about them without ever looking at the source of a Chess engine, by just browsing through forum postsor Chessblogs. As indeed I did before writing micro-Max.

[marcelk]

Or are the programmers today more intelligence as in the time we powered Winboard at first?

I don't think this is a mystery at all. Before the CCRL existed there obviously was noCCRL rating scale, so the ratings you quote must heve been from another scale, which makes it doubtful that they can be compared to CCRL ratings at all. If the 'unbeatable 2500' was relating to a Human GM rating of 2500, hardware is not factored out. On 20-year-old hardware Houdini would of course also not be above 2500 Elo.

And then there are of course the 'open ideas' that every one uses, and which give a huge Elo boost to simple programs: alpha-beta, QS, MVV/LVA move sorting, null move, LMR, hash table, killer heuristic, mobility eval. Implement all that correctly, and you are already at 2500 Elo CCRL. Micro-Max, with only 100 lines of code, only implements the most necessary part of that (no real move sorting, no killer, no mobility, no passer evaluation, no piece-square tables), and it is already around 2050 Elo CCRL.

Also don't forget the impact of elostat and multicores. These mark the transition from 'alchemy' to 'engineering' in computer chess.

[Frank Quisinsky]

Hi,

yes, that could be or is a good explanation. Logical for myself what your wrote.

On the other hand ...
So many engines started with many of ELOs more as for around 10 years. What I mean is, not easy to say ... all this engines are clones if the programmers used good known ideas which are secret for 10 years.

Very complicated for a NON programmer like me. What you wrote is logical for me but perhaps who know it in detail.

We can give Vas a yellow card.
We can give him not a red card.

That's my opinion!
And I am sure if we give Vas a yellow card much other must get this card too. And each programmer which signature the "Open Letter" have to give the own sources for a check before an official tourney will be start.

Best
Frank

[marcelk]

And each programmer which signature the "Open Letter" have to give the own sources for a check before an official tourney will be start.

I would support any initiative in which investigations are not started after suspicion is raised and a formal complaint is filed, but as standard operating procedure. Investigating all programs is possibly not feasible, we don't have Formula1 like budgets. Random checks might be. Or always checking the winner. There could be several ways to work out the details that are acceptable by all.