Author: Robert Hyatt
Date: 12:29:26 07/21/00
Go up one level in this thread
On July 21, 2000 at 14:48:06, Ed Schröder wrote: >On July 21, 2000 at 10:58:47, Robert Hyatt wrote: > >>On July 21, 2000 at 03:08:09, Ed Schröder wrote: >> >>>On July 20, 2000 at 20:54:31, Robert Hyatt wrote: >>> >>>>On July 20, 2000 at 15:38:10, Ed Schröder wrote: >>>> >>>>> >>>>>Yeah..... let's talk about chess! DT losing in Hong Kong 1995 and never >>>>>trying to get the world champion champion title when they had the chance >>>>>to proof that Hong Kong was a mistake. Perhaps it was no mistake? >>>>> >>>> >>>> >>>>OK... Fair enough. How many WCCC events or WMCCC events did _you_ skip? >>>>Why? I believe you said you got tired of the book wars. But you could >>>>have still competed, right? And maybe done poorly if your book was busted? >>>>And sales would have suffered? >>> >>>The list of main events Rebel participated can be found on: >>> >>>http://www.rebel.nl/r8-r9.htm >>> >>> >>>>Remember that after the Kasparov match, IBM wasn't going to do _anything_ to >>>>taint the incredible public relations coup they obtained by beating the world's >>>>best chess player. The marketing guys would have gone ballistic had Tan allowed >>>>DB to enter a computer chess event where any possible outcome except for 100% >>>>wins would have resulted in lots of chess-thumping "I beat Deep Blue..." >>> >>>This is not correct. We were promised a DB Internet version everybody >>>could play. Where is it? >>> >> >>You have to sit back and think about the following: >> >>(1) Hsu was _always_ open. He competed in nearly every computer chess event >>he could, as did I. He and I are very similar in several respects, one of which >>is that we both like to compete with our "creations". >> >>I _do_ compete all the time. I don't make the WMCCC event because it is too >>expensive, and it comes right in the middle of a term where I am teaching and >>missing over a week of classes is not reasonable, IMHO. But I _do_ play on ICC >>all the time, where anybody can play or watch. >> >>If Hsu had the choice, he would too. Do you know the "thing" with the best-ever >>win/lose ratio on ICC? Hint: Scratchy -- the handle for deep thought. >> >>(2) Hsu worked for the largest computer company on the planet. They didn't >>work for him, he worked for them. Along with all the agreements an employee >>must make about intellectual property rights, and so forth. And when he beat >>Kasparov, the marketing division was ecstatic. And the marketing division >>_runs_ IBM, as it should. And _they_ made the decisions about what could and >>could not be done after the match. From a purely business perspective, it makes >>_perfect_ sense to not play again. Because there is _nothing_ they could do but >>do worse than they did against Kasparov. >> >>To imply that it was Hsu/Campbell/etc is 100% incorrect. They worked for IBM >>and were obligated to do what they were told. Whether we like it or not. I'd >>love to have DB on the internet. But it isn't our decision. Nor Hsu's... > >The DB Internet version was a promise, did you see it? > >My point was about IBM not fulfilling one of the many promises they did. > > Let's use a biblical expression... "In the beginning, the DB project at IBM was just another special-purpose hardware solution project." Hsu could have promised anything he wanted, and might have even gotten his manager's approval. But then things changed. The first match caused a huge P/R uproar and the marketing guys took notice _real quick_. And from that point forward, even according to Hsu's accounts, he "lost control". I could promise to let you use my quad xeon in a chess tournament, just as I have done for Vincent several times. And someone higher-up here at UAB could say "no" and "no" it would be. Beyond my control. > > >>> >>>>I don't what happened. But it was certainly predictable. I have no doubt >>>>that they would be an overwhelming favorite in any computer chess event. But >>>>Hong Kong can happen again. All it takes is a communication failure and a >>>>restart at the right instant and <blam> you play a bad move. Hong Kong proved >>>>it _could_ happen. >>> >>>>And IBM marketing would _not_ allow that chance to be taken. >>> >>>A little bird told you or IBM? >> >>I big bird at the top of the DB project... > >Who? > >Anyway it contradicts a promise made. > In business, a "promise" is far less important than "a signed contract" which is what IBM had from Hsu. :) > > >> >> >>> >>> >>>>From a business perspective, they would be utterly stupid to play in any other >>>>event, until the long-term 'buzz' from the 1997 match fades into the past... >>> >>>If you think you are so superior (as they claim!) I would like to show to the >>>world. I also believe that if you make such claims you are obliged to proof >>>it. >> >>If your boss says "NO", he probably means "NO". Would _you_ defy management, >>go out and play in a tournament and lose one of those odd games, and then have >>to face the legal action where IBM could _prove_ you cost them millions of >>advertising dollars? Be realistic. In _any_ company like that... IBM... >>Cray... SGI... DEC... the marketing guys _rule_. >> >>It is good business for it to be so. Because the goal is to sell machines, >>not advance science. Even if the goal of the individual employees is not the >>same. > >200M NPS is good for business too. > >Maybe it was (is) only 10 million? > >Looking at the number of unfulfilled promises I don't take everything >for granted from IBM, not anymore. > I think that line of reasoning is silly and doesn't require a response. We know how fast it was _before_ Hsu left CMU? Or did the conspiracy with IBM date back to a time before they had any idea IBM might hire them??? IE back to 1987 when it first started playing? > > >> >>> >>>The IBM pages are full of claims, here is one: >>> >>> "Over the years, Chiptest evolved first into Deep Thought, then >>> into Deep Blue, the most powerful chess-playing computer ever >>> constructed." >>> >>>This was written in 1997 while another program was world-champion >>>in that period (1995-1999) nota bene beaten in a direct confrontation. >>>I call this kind of information misleading, softly speaking. >> >> >>It isn't misleading at all. It was certainly factual. Could Fritz search >>a minimum of 200M and a max of 1000M nodes per second? If not, then IBM could, >>and clearly that was the most powerful machine around that played chess. Power >>doesn't mean "best chess player" although they could easily make that claim as >>well and experts in the field wouldn't dispute it. > >You can't say you are the best having lost the World Championship. IBM >had the courage to do. And you find this not misleading? No. Which says more about a tennis player? Winning Wimbledon, or being ranked number 1 in the world? There was no doubt DB was ranked number one in the world, if you took a poll of everyone qualified to make that judgement. > > > >>> >>>Tell me why should I believe the IBM propaganda and everything else >>>they say. Heck I even have started to doubt the never questioned 200 >>>million NPS as just being good for their sales. >> >> >>Doubt what you like. Just notice their search depth of 17-19 plies in the >>middlegame and ask how they did that if they weren't going at 200M nodes per >>second and beyond... > >A couple of points to the above.. > >1) You say 17-19 plies in the middle game. Each posting of yours I see >these numbers automatically increased. What is next? 18-20 plies? I see numbers from 10(6), 10(7) to 12(7) in the middlegame. You pick the number you like. 16-18 is low. 17-19 is more common. I changed the number slightly as I have been going thru the logs as you asked, remember? You can go thru them yourself and decide which number you like and report it here. You simply add the two numbers together. The first is the software search depth, the second is the hardware search depth. The sum is the total search depth... I'll let you pick a number that represents the log results since you don't seem to like mine... > >2) DB is no brute force program (as you always have claimed). Quote >from the IBM site: > > "Instead of attempting to conduct an exhaustive "brute force" > search into every possible position, Deep Blue selectively > chooses distinct paths to follow, eliminating irrelevant searches > in the process." > >I always said this after I had seen the log-files. It beats me how you >always have claimed the opposite on such a crucial matter presenting >yourself as the spokesman of Hsu even saying things on behalf of Hsu >and now being wrong on this crucial matter? Sorry, but you are wrong and are interpreting that wrong. DB uses _no_ forward pruning of any kind, this _direct_ from the DB team. The above is referring to their search _extensions_ that probe many lines way more deeply than others. If you want to call extensions a form of selective search, that is ok. It doesn't meet the definition used in AI literature of course, where it means taking a list of moves and discarding some without searching them at all. > > >>The hardware details have been described in various journals. 480 * 2.2M nodes >>per chip says (to me) that they _really_ could have claimed a much bigger >>number and been just as honest as any other SMP programmer around. They claimed >>only 20% of their peak number, where everyone else reports their peak... >> >>Doesn't sound exaggerated to me at all... >> >> >> >> >> >>> >>>IBM has been caught on lies and false interpretations (the match being >>>scientific, promising Kasparov to give him full explanations after the >>>match) and more of such. Why should I trust any information that comes >>>from a source that has proven itself being unreliable providing misleading >>>information. >>> >> >> >>Were the log files not released eventually? Would _you_ give Kasparov >>_anything_ after the way he behaved during and after the match? I wouldn't... >> >> >> >> >> >> >>> >>> >>>>>What about DT not seeing a simple tactics on tournament time control (!!) >>>>>every chess program sees within 10 seconds? >>>> >>>>You do remember Hsu's explanation? That DT _had_ found the right move in >>>>that bad book line (g3 I think, I am not sure). And a communication failure >>>>caused them to restart and it moved before it saw the problem with the move >>>>it played? That has happened to me. It is part of the game. And they lost >>>>because of it... >>> >>>Any chance you can back this up? >>> >>>Are you suggesting 16. c4????? came from a hardware or communication >>>failure? >> >> >>Hsu reported this himself. He said that DT had liked c4 for a while, then >>changed away from it, then came back to it, then failed low and found g3. >>They had some sort of network/communication failure while it was still Fritz's >>move. They discovered this apparently when Fritz moved and they couldn't talk >>to their program to enter the move. They reconnected, but were not using a >>client program that would keep DT running. So on the comm failure, the >>program was terminated. They restarted the program, and it moved pretty >>quickly. >> >>You can certainly email Hsu or Campbell and ask about the details. I think >>this was discussed in r.g.c right after the tournament, but I am not sure. >>It might have been mentioned in his upcoming book as well, again I am not >>sure and don't want to re-read the whole thing to look. > >Whatever they say I don't believe this. When you do 7 million nodes per >second 16.c4?? should be refuted within half a second. When this is not >the case this says enough about the quality of the program. This _was_ deep thought. It was doing about 2M nodes per second in 1995, according to Hsu. Whether it overlooked the tactic, or the extensions made it think the move was not so bad at first, is impossible to say. As I said, you played f4 and blew a game badly in your GM challenge. Other programs saw f4 losing quickly. Does that mean yours is weak? or just a bad eval term acting up? They can do the same of course... > >It's either one way or the other and 16.c4?? proofs it. No excuses. Just >caught and no escapes. OK. DT was way overrated, because it made this bad move. Of course, Rebel also sucks as it played an easily identifiable lemon move f4 vs a GM. > >BTW: your Crafty rejects 16.c4?? within a couple of seconds. NPS about >250,000. Why can't a machine that is 28 times faster in NPS not see this >nota bene on tournament time control? > > Is it not possible that _their_ evaluation was different and didn't think c4 was nearly as bad as we do, based on some correct (or incorrect) evaluation term that was bigger/smaller than ours? I make plenty of silly moves. So does Rebel. Junior. Fritz. Etc. Does that mean they are weaker than it seems? We can make mistakes and it is OK, but deep thought has to play perfect or it is vastly over-rated??? > >>> >>>DT lost because of missing a simple tactics, just 10 plies deep >>>and that on tournament time control while every program will see >>>in a few seconds. Even after a restart 16.c4?? should have been >>>rejected within a few second **IF** the machine is the tactical >>>beast you want us to believe. >> >>If it missed simple tactics, wouldn't it have lost games right and left? >>I can think of several reasons why they might take a while to see that c4 >>is bad... a program bug -- we all have 'em. An operator error on the >>restart. This happened to me in Jakarta where someone pulled the plug on >>our machine in the middle of a game, and when the thing was restarted, the >>time control was entered wrongly (I wasn't there). I lost round one because >>the operator blew the time control thinking he was talking about 40 moves in >>90 minutes, but he entered 40 moves in 90 hours. There are too many things >>that could have gone wrong in that game right after a restart... But since >>that was one of two games they lost over quite a period of time, I don't >>conclude they were tactically missing things right and left. On the contrary, >>they were generally seeing things (at all the ACM events I attended) that were >>leaving their opponents and spectators gasping... >> >>Again, ask Mike Valvo. He was the TD at all the recent ACM events. Maybe you >>will believe him? >> >> >> >> >> >> >>> >>>So now we have 2 cases from practise DT/DB being caught not being the >>>superior beast followed by 2 explanations from the IBM camp (bug, hard- >>>ware failure) assuming your informations are correct. >>> >>>Especially the 16.c4??? excuse is an extremely poor one given the fact >>>16.c4 is about a very simple tactics. >>> >> >> >> >>Being the best, and being "perfect" are not the same thing. I think it >>funny that one bad move, in one very rare game, is used to mark DT as >>inferior. In spite of the 50 other victories against the best computer >>programs in the world. In spite of their having a trophy case so full >>it is bursting at the seams... > >This item has been successfully refuted by Chris in previous postings. >It was about winning from 5 Mhz 6502 and 386/486 machines. Big deal. >When the P90 came the machine was in trouble. You really believe this? We had P5/133's at the last ACM event. DT had to forfeit round 1 due to a power failure at their lab. They _still_ won the event, and beat machines _much_ more powerful than a simple P5/133. IE *socrates on a 512 node CM5. Cray Blitz on a C90. I don't remember who else (ah yes, Mchess Pro was one of the last two) but I can look up the exact results if you want... You _really_ believe that a P5/90 was a problem for deep thought. _really_? Is this a statement you would make at a major computer science conference with hardware and software people filling an auditorium? :) > > >> >> >> >> >> >>> >>>>But remember, that was one of exactly two games they lost to a micro program >>>>since 1988. Pretty tough to follow such a dominating performance... >>> >>>This argument has been already successfully weakened by Chris. DT winning >>>from 5 Mhz 6502 and 386/486 machines is not exactly something to be proud >>>on. When it had to face a simple Pentium-90 things went wrong. After that >>>they just disappeared still claiming being the best. Not very conving. >>> >> >> >>As I mentioned before. DT evolved each year. Remember the speed analysis >>I gave about how they had gained over a factor of 3.3 against rebel since >>they first came on the scene. And that includes 3 years of inactivity by >>them, while we continue to get faster and faster hardware each year. And >>they are still 3.3x faster when you compare 1988 DT to 1988 Rebel, vs 1997 DB >>to 2000 Rebel. It will take 2 more years for you to get back to the same >>speed ratio with them you had in 1988. If Hsu did his new chip, he would gain >>about 30X more, which would mean he has widened the gap from 1988 to today >>by a factor of 100... Falling behind by a factor of 100 in 12 years is pretty >>bad. I can extrapolate what those two curves are doing, and it looks bad for >>the one getting left behind. Note that _I_ am behind too... > >I do take the 3 games into account where DB Junior lost to Rebel >with 3-0 and the 3 games DB Junior played against Chess Tiger >which ended in 1.5-1.5 Chess Tiger playing on a poor 150 Mhz laptop. > >And I don't believe the explanations given by Hsu in public on these >6 games being just a 2200 elo program as in private email between >him and me there is contradiction and I come to another conclusion. > >The thing Christophe and I played in Paderborn last year was DB-JR >just as the documentation said on the screen. No single statement >seen this was a crippled (demo) version (not on the screen nor in the >documentation that came with the program). > Fine. Again, Hsu is a liar. If that is what you want to think. Here is an excerpt from him that might help: =============================================================================== Web-based DB Jr uses a single card, a random opening book (including fairly bad lines) and one second per move (a quarter of which is used in downloading the evaluation function, and the search extensions are more or less off due to the very short time). It probably plays at around 2200, which is usually sufficient to play against players in random marketing events. Repetition detection is also turned off (The web-based program is stateless). The playing strength of "DB Jr." spans a quite wide range, depending on the setup. The top level, which we used for analysis and in-house training against Grandmasters, is likely in the top 10 of the world. ================================================================================ Now tell me how there is a contradiction there and how I am misinterpreting what is being said??? Seems _perfectly_ clear to me... > > > >>> >>> >>>>> >>>>>What about the DB-GK position Uri posted recently DB being dead wrong >>>>>not seeing a giant material loss? >>>> >>>> >>>>The fishy PV problem? That is common in their output and doesn't bother me >>>>a bit. They can't get the PV like we do, so they have to probe around in 32 >>>>processors to get the various "best moves". And they occasionally get nonsense, >>>>which is not totally unexpected... >>> >>>Nope. I am talking about Rd1?? see my other posting. >> >> >>I thought you or someone else had reported that they had a gross bug that was >>found after that move was played. Dealing with a C macro that expanded in a >>way that badly broke something in a particular circumstance? >> >>Again, one bad move and we dismiss DB? >> >>In that case, we must have dismissed Rebel, Crafty, Deep Junior, etc. _years_ >>ago because they all make _horrible_ moves. Remember "f4"??? :) It happens. >>It doesn't mean the program is weak. > >The 2 positions in question reveal quite a lot. It tells you something >about the tactical power of the program something that is quite dominant >in computer chess. You as defender of search are the first one to know >that. Two positions do not an engine make. Otherwise the best player in the world is a patzer. He did make the worst possible move in a drawn game and resign, right? DB made a bad move in one game vs Kasparov and lost. It was lost in any case, so it really had no effect on the game. > > > > >> >>> >>> >>>>Didn't bother me as the score was reasonable, as was the move they played in >>>>the game. >>>> >>>> >>>>> >>>>>What about the 3 games Chess Tiger played last year in Paderborn against >>>>>the Internet version of DB-JR versus Tiger running on a slow 150 Mhz? In >>>>>case you forgot the score was 1.5-1.5 >>>>> >>>> >>>> >>>>Against a crippled version using almost no time to search, with no repetition, >>>>no 'state' of the game, etc? I get thumped all the time on ICC when I run up >>>>a new version with a serious glitch. Or when something else is running so that >>>>I get 1% of my CPU for a couple of moves. Does that mean my program is weak??? >>>>On the quad??? >>>> >>>> >>>> >>>> >>>>>Not to speak about the 3-0 Rebel scored against this DB-JR Rebel running >>>>>on a simple 333 Mhz notebook. At least these games were real, real in the >>>>>sense 6 games were published and many people have watched them live. I was >>>>>not shouting 3-0 only at least I could produce the evidence. How about >>>>>these supposed 40 games? I have never seen one. >>>>> >>>>>Well... this is what you get when you hide, do not play, shout 36-4 and >>>>>provide no evidence. >>>>> >>>>>Ed >>>> >>>> >>>> >>>>They didn't hide from 1988 to 1995. Where were you then??? >>> >>>- 1991 Vancouver: world champion micro's >>>- 1992 Madrid: world champion all classes >>>- Various first places on SSDF >>>- Overall best computer at AEGON (man vs machine) >>> >>>Good enough ??????? >>> >>>Ed >> >> >>You miss the point. They weren't hiding? Where were you then? Didn't you >>see them at each of the ACM events? Or see their results? Did you see them >>at all but the 1992 WCCC event? I wasn't asking where _you_ were competing. >>I was asking where _you_ were when they were competing at every computer chess >>event they qualified for. I don't think you can say they were "hiding" when >>they were at every ACM event I attended, and I didn't miss any that I can >>recall. They were at the 1989 WCCC event in Alberta and at the 1995 WCCC >>event in Hong Kong. Hard to say they were hiding when they attended more >>events than anybody else... > >I said they are hiding after the Hong Kong 1995 debacle, not before. >7 million NPS versus a couple of poor P90's and the dream was over. > >Ed IBM might be hiding, to be sure. They did something no one else has done, and which no one else is likely to do in the near future. After beating Kasparov, what more _could_ they do? Any loss or draw of any kind would mar that marketing gem. I don't like the decision, but I would probably make it if I was VP for marketing at IBM myself. My goal there would be to extract every dollar of P/R from that match that I could. They are _still_ reaping benefits from that today, 3 years later. I don't think we can blame them for that. At least the shareholders would say they did the right thing, whether we agree or not...
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.