Slashdot Mirror

In fact, you are wrong on several points.

On which points?

First, the strongest program is Stockfish 6. It's still improving at a rate of 50 ELO points at each version, and is already is above Komodo:
http://www.inwoba.de/
You can see that Stockfish 6 is already 200 points above Rybka.
Stockfish is improved by a community and by using a distributed network: http://tests.stockfishchess.or...
The current version is already stronger than SF6.

OP gave a brief intro and history of 3 programs. He said that Rybka took an early lead in strength, but then others have since improved on that. Your point about SF6 being currently stronger than Rybka may be true, but it doesn't invalidate anything that the OP said.

Secondly, Rybka has been demonstrated as a copy of Fruit (an open-source chess engine), with only bit-tables added.
There has been an incredibly detailed decompilation about Rybka http://www.chessvibes.com/plaa... which leads no doubt about this.
The only difference in recent versions of Rybka is that the evaluation function has been improved by GM Larry Kaufman, but he works now on Komodo.

I have no doubt that Stockfish is stronger than Carlsen, except that it does not use a creative style.

OP made no claim about the source of Rybka. He said Houdini was built on Rybka.

Your comments are informative and add value to the conversation, but your opening statement saying the OP was wrong on several points doesn't seem to have any merit.

In fact, you are wrong on several points.

First, the strongest program is Stockfish 6. It's still improving at a rate of 50 ELO points at each version, and is already is above Komodo:
http://www.inwoba.de/
You can see that Stockfish 6 is already 200 points above Rybka.
Stockfish is improved by a community and by using a distributed network: http://tests.stockfishchess.or...
The current version is already stronger than SF6.

Secondly, Rybka has been demonstrated as a copy of Fruit (an open-source chess engine), with only bit-tables added.
There has been an incredibly detailed decompilation about Rybka http://www.chessvibes.com/plaa... which leads no doubt about this.
The only difference in recent versions of Rybka is that the evaluation function has been improved by GM Larry Kaufman, but he works now on Komodo.

I have no doubt that Stockfish is stronger than Carlsen, except that it does not use a creative style.

I also play chess at tournament level and respectfully disagree. Many grandmasters make moves that are unintuitive, such as Fischer, and I do not see how you could distinguish a very good computer's move from a grandmaster. I analyzed the game here: http://www.chessvibes.com/reports/bulgarian-chess-player-strip-searched-after-suspection-of-cheating and could point out some interesting moves, but nothing that rang out as a computer move.

From my own experience playing chess online, I end up losing on purpose to see if a person is cheating with a program. I'll allow fools mate to happen to me twice in a row, where any normal chess player would let it pass the second time, the chess program will always take the quickest mate.

The submitter should have linked to this article directly. It gives a lot more detail, and at the bottom links to the actual evidence (comparisons of disassemblies with Crafty and Fruit, the reports of the experts, and such).

First of all, the guy who wrote the article is just clueless. No offense to him, but he sounded as one who has clearly no competence to speak on the matter.

Secondly, the panel only made a recommendation. The ICGA board plus WCCC tournament director acted like the jury. They agreed with the recommendation.

Thirdly, Ken Thompson was one of the members of the panel and I quote investigation report:

Here are some opinions of panel members recorded on the investigation wiki pages:
Ken Thompson:
After reading all the evidence, voted that he was convinced by the case against Rybka.

If you are going to call him an asshole, I'm definitely going to call you an idiot.

ICGA didn't f*ck up.

As posted previous -- here is the PDF of the report: http://www.chessvibes.com/plaatjes/rybkaevidence/Rybka_Investigation_report.pdf

Specifically see section 2. Investigation:

2.1 Executable file analysis

Quote “Rybka 1.0 Beta and Fruit 2.1 have exactly the same evaluation features“. Disassembly of the root search analysis indicates nearly identical code and variables, even including the ordering of the variables. Appendix B on the evaluation of Rybka 2.3.2a shows “the evaluation function in Rybka 2.3.2a is substantially the same as in Rybka 1.0 Beta”. Watkins compares evaluation function features between Rybka, Fruit and four other open source programs (Phalanx, Pepito, Crafty and Faile). In section D.2.3 he shows an overlap of 18.8 out of 24 evaluation terms for Rybka-Fruit (73% overlap), with much lower overlaps than between the other programs. Crafty 19.0-Fruit eval overlap: 12.9/24 = 54%, Phalanx-Fruit overlap: 43%, Pepito-Fruit overlap 37%, Faile-Fruit Overlap 23%. This has been expanded with more statistical rigour in a separate 50+ page paper mentioned in 2.4 below.

Why the hell do things like this bring out all the Slashdot loonies? You people could have read the article, but you don't even have to do that. canajin56 has done it for you. They didn't just compare some moves. They actually analysed the binaries. Their case is 100% watertight. The only parallel I see with the DHS is you and a bunch of other Slashdot posters accusing them of ignorance before you have even read anything about their report.

Maybe you should read the ICGA's decision.

At the bottom of that page, you can also find the evidence that they used to reach that decision. To any programmer who can read a disassembly, its pretty clear that the eval in Rybka 1.0 Beta was derived from Fruit 2.1, and this situation persisted until some version later than Rybka 2.3.2a (probably around the time of Rybka 3 when the eval was rewritten).

Or you can talk out of your ass some more, that works too.

The problem with looking at just algorithm similarities is that every modern chess bot uses some variant of the same algorithm so the executable code will superficially look similar (negamax.) Computers aren't powerful enough to search the entire game tree, so you have to stop after a certain number of levels (15, for instance) and use a heuristic to evaluate the strength of that position.

The main differences between chess bots are found in that heuristic.

According to the actual report the heuristic is obviously based on Fruit's, which is what they're really angry about.

Fruit was a GPLed program until and including version 2.1. The investigation panel had compared Rybka 1.0 beta to Fruit 2.1, then they took a look at previous and later versions, including Rybka 2.3.2a to establish a pattern behavior. You can read all that has been done in the report: http://www.chessvibes.com/plaatjes/rybkaevidence/Rybka_Investigation_report.pdf