BB's Rybka/Ippolit comparison

[BTO7]

A thread about this thread was started in rybka forum...but has suddenly been moved or deleted hmmm. That looks like the big NO comment hurry get that out of the public eye as fast as you can demonstration to me

BT

[Jeremy Bernstein]

The timing of this document could not be more perfect considering the email exchange between Vas and Sven, see CCC.

The chaos is complete.

Ed

I have some reservations, however, I think there is no coincidence that they occured so close to each other. They waited for Vas to play his hand, then responded. Maybe ?

Zach offered to post this paper at least a week ago. I think he was just waiting for permission from BB to do so. I think it's just a coincidence.

Jeremy

[Angel]

Weve posted the report on Ippolit Wikie and our own forum ,so its not going to go away as Vas and rykba co might wish ,,they can fool some of there forum members all of the time and most some of the time but not all ,all of the time

[kingliveson]

At least we can expect to get response from Vas in 2 years.

[Gino]

Very interesting paper for a lay person like myself. If I was a jury in a potential lawsuit, this would be a very convincing argument as long as other programmers would agree with the analysis presented here. My only doubts would be related to reasons why this clever programmer remained anonymous?
Someone mentioned this person could have been part of the rybka team?
Even if we accept the programmer created the code, if he had been a member of the rybka team then maybe he could not use ideas obtained from rybka either. I know this would be the case in other fields of science but maybe they don't apply here.

[BB+]

I can't say that I am eager to delve into a discussion right now, but I will make some exterior comments:

just by using the output of objdump.

Not exactly (objdump requires an ELF format, no?), but I take it that "disassemble" in gdb (with Linux/microwine) is just as caveman. Following the general structure of the IPPOLIT code can help somewhat. I also wrote some code, which in conjunction with breakpoints in gdb, allowed more assurance in some cases.

I'm assuming the comparisons and detail are all accurate

I don't claim every thing is 100% accurate, and in fact, when I posted it on Zach's secret forum, I specifically included "corrections and" when saying "comments are welcome". By and large, I think any errors are sufficiently minor so as not to overturn any conclusion [I might mis-speak on some of the Rybka bits, particularly where I had to figure out what was happening (with no IPPOLIT version as a crutch), and I didn't cross-check every last detail with IPPOLIT against IPPOLIT.C or IPP_ENG.C in the end]. Of course, evidence is always open to reconsideration, if say, someone were to offer an alternative narrative of what is seen. One man's analysis is really only a place to start a synthesis. Other than GCP, I'm not sure that anyone else has commented on Rybka/IPPOLIT differences at much of a technical level.

there is the possibility that the Rybka code and the IPPOLIT code were written substantially by the same person/team, if I was investigating this, that's were I would put a strong line of questioning and research

I disagree with this. Why bother making so many changes if so, such as reordering the pieces, changing hash entries from 64-bit to 128-bit, etc. My take on the "goal" of IPPOLIT was to write something that was substantially un-Rybka-like if one investigated it with a fine-tooth comb, but to release it such a way that made it "obvious" that it was a clone. [Why, for instance, did IPPOLIT go unmentioned from May to October --- could the programmers not find a forum for it? Well, maybe the authors gave up, as everyone thought they were kooks akin to "Ben Lau", but more likely they wished to play-up the "censorship" angle -- but I would rather stick with facts, as opposed to psychological speculation].

if I were an investigator, "BB" would be a suspect on my list.

Perhaps I should have been hurt if I were not... [After all, the list of known and competent suspects has got to be rather small, maybe no more than 10 or 20].

Is there a BB Rybka/Fruit report?

This is in the works, though I would be borrowing a lot from Zach at present. In fact, I had wanted to address this issue first, but c'est la vie (or "se la vi" as I once read as a student excuse when grading homework). Also, Zach and I do not always agree on the relative turpitude of various similarities.

I'm pretty sure Felix would have started a lynch campaign on him on Rybka forum long time

I gave up posting on Rybka forum a few months ago, due largely to the annoyance over the moving of threads. I had been variously absent for multiple months at previous times, so this is not completely aberrant. I don't think Felix is exactly to blame (the worst I had with him was questions as to why he didn't suppress other posts mentioning clones, and sometimes he agreed to move them), and he certainly never started a campaign against me, nor was I banned.

he actually mentioned using "painstaking analysis via debugging tools", whether it means the same as others talk about decompiling.

There is a technical difference (in my understanding/nomenclature) between disassembly (using debugging tools) and decompiling, though from my standpoint, "all code is source code" in some sense. With decompiling, I would say that code in the target higher-level language is automatically produced (and typically quite messy), while this is not true with disassembly. However, if you aim to "understand" the code in any case, I don't see why a screw-ball high-level intermediate form is really too useful (breakpoints in gdb are nearly as good as compilable code).

The timing of this document could not be more perfect considering the email exchange between Vas and Sven, see CCC. The chaos is complete.

Yes, the timing of certain events has always been a background issue involving these "clones" (for instance "Yakov" seems to have first appeared about a week after Jury Osipov claimed in a Russian forum to have an engine that was 50 ELO better than Rybka 3 -- and then IPPOLIT went viral about 2 days after "Yakov" started speaking "Spanish") -- however, in this case, I finished that document in mid-May (about six months of on/off --mostly off-- preparation), and posted it on Zach's secret forum. He took a couple of weeks to read it (busy with WCCC -- he had told me back in March of his Zappa connection), and then asked for permission to post it elsewhere. This was already done a few weeks ago; Zach can speak for himself, but my impression is that he was just busy with other things between when he offered to post it on TalkChess and when it appeared here. Of course, to the conspiracy theorist, exploiting coincidences to their maximal impact is a known political technique.

Incredibly detailed proof that Ippolit is not a reverse-engineered Rybka!

Depends on what you mean by "reverse-engineered"... I would find it wholly plausible that IPPOLIT is a "reverse-engineered Rybka with many changes", and indeed I would assert that the IPPOLIT maker(s) certainly knew much of the internal workings of Rybka. It is not, however, merely a "code-based" copy therein (and my recollection is that the claim, at least at one point, was that code of Rybka appeared in IPPOLIT). I see no way to differentiate as to whether IPPOLIT "started from scratch" and implemented Rybka-like-parts one-by-one, or "started as a functional equivalent to Rybka" and then was modified (rather substantially). There are a few clues, for instance IPPOLIT computes Crafty-style bitboards at startup (as opposed to just having huge arrays in the executable), to indicate that at least some parts of IPPOLIT are completely independent (code-wise) from Rybka.

One thing I always asked, that people needed to look at RobboBase tablebases implementation -- they know what they are doing.

My interest in bitbases (or "TripleBases") led me to look at this too. There are some interesting ideas, but I think it is largely what was already known (both in the checkers literature and, presumably, with ShredderBases). They (or "Roberto Pescatore") have some new ideas, but most of what is there is not exactly earth-shattering. I do agree that these style of "bitbases" (run-length-encoding) are Best Current Practise in the field. The "TotalBases" are just DTC tablebases. They are slightly different from Nalimov, in that they are "one-directional" (as with FEG, only having is-win, is-draw, lose-in-X), and use Burrows-Wheeler compression, as the Huffman-only idea of Nalimov (to have fast realtime access) is presumably not necessary if you can have the bitbases immediately accessible from memory. Again a lot of this is standard, though it does save a lot of disk space. Negative aspects of the code are that it doesn't build in parallel (though it is largely I/O bound in any case, at least for 6 pieces, unless you have around 32GB and can do this in memory), the disk build (for 6 pieces) is a bit flaky, it is "only" DTC, and the compress routine looks rather slow (building a suffix tree -- admittedly, compression need only be done once per tablebase, and their decompression actually looks a bit faster than bzip2, possibly because they seem to obviate the need of using and switching between multiple coding tables).

Btw. anyone knows who is the guy?

Unlike others, I prefer to be anonymous in general fora. Contrary to what others find, for me not being able to put a face on another person makes me less likely to indulge in personal attacks, and similarly being anonymous reminds me that I should sculpt my argument with facts rather than at personalities (for instance, I don't think I ever "blew my steam" on the Rybka forum, until Felix and Soren started being a bit pushy on "cloning" posts). That being said, I've given numerous hints on the Rybka forum (I think Nelson claimed that one could "google me down") to those interested in establishing my identity. Also, on Zach's secret forum, mostly due to its smaller size and communal aspect (the only thing akin to a "personal attack" I saw there was by VD on someone not on the forum). I chose to use my real name, I think Zach is the only person I personally know on this forum (I met him in March), but my history in the subject is long; I was testing Crafty (rather mindlessly on test suites -- the university had a SPARC, and I used it) back in the 90s, had some email correspondence with Hyatt a few years later, then changed to writing other games programmes for a bit, became otherwise involved, and didn't come back to computer chess until maybe 2006, and still only pursue it as a hobby.

[mcostalba]

but c'est la vie (or "se la vi" as I once read as a student excuse when grading homework).

Are you italian ?

That phonetic "translation" can be written by an italian, for sure not by an english speaker and of course not by a French

I'm not sure how german people would write c'est la vie in their language.....

[BB+]

That phonetic "translation" can be written by an italian, for sure not by an english speaker
No, this was in the southern US, when I was a grad student. I'm guessing the student (also very Southern) thought the phrase was Spanish. Reticules... [See http://extempore.livejournal.com/198866.html for the context].

[thorstenczub]

The timing of this document could not be more perfect considering the email exchange between Vas and Sven, see CCC.

The chaos is complete.

Ed

any lie comes out
one day.

now this explains why there was no rybka3 update. because vas had no rights on the sources anymore when the programmer who had done it, left the company in an argument.
this would explain why this programmer felt to damage vas sales by putting ippolit online.

[Matthias Gemuh]

but c'est la vie (or "se la vi" as I once read as a student excuse when grading homework).

Are you italian ?

That phonetic "translation" can be written by an italian, for sure not by an english speaker and of course not by a French

I'm not sure how german people would write c'est la vie in their language.....

Sounds odd, but there is no phonetic equivalent of "c'est" in German. Then word would have to start with "Ss" and that is not allowed.

Matthias.