Author: Rolf Tueschen
Date: 17:41:44 08/25/02
Go up one level in this thread
Let me just thank you for the - I take the word - _beautiful_ explanations. I can only hope that you also enjoyed my questions. Since I often concentrate on such topics where these famous persons are involved, I want to explain why this is the case. It is not at all, because I search for such figures to then being able to dump them or also to let fall some light from them on me. No, for me as a real layman in the technology of CC it is highly inspiring to try my own thoughts from totally different fields, mostly psychology but also the science issues as such, and finally statistics. Of course questions in other topics, where comparably unknown experts are involved, could well be even more fruitful viewed from a pure optimal thinking. But then I must admit that since I'm not able to understand the _real_ importance of questions, I am condemned to follow the more primitive way, and that is the one over names! I think that is similar to many here who might read your particular posts with preference and leave posts aside that are perhaps much more important - objectively. Hope that people will believe me. The observation is right, names are the highlights of attraction, but my motivation is _not_ going after names to bash the individuals behind them. Finally with you, I am always very happy if I could get something out of you just like that here in your last posting. Learning something is always much fun. I just came home in the middle of the night here and this was very nice reading. I'm still reading the biography of John Nash in daily bits while resting on the balcony, I'm now now his 1970s, when he was the so-called ghoast at Princeton. Isn't it good that we can have so much fun with these talkings?! In the international community. Excuse the emotions. I think this was once necessary after so many prejudices, conflicts and misunderstandings. Rolf Tueschen On August 25, 2002 at 11:17:01, Robert Hyatt wrote: >On August 25, 2002 at 06:36:19, Rolf Tueschen wrote: > >>On August 25, 2002 at 00:24:02, Robert Hyatt wrote: >> >>>On August 24, 2002 at 18:23:14, Keith Ian Price wrote: >>> >>>>On August 22, 2002 at 22:35:25, Peter Hegger wrote: >>>> >>>>>On August 22, 2002 at 18:19:17, Keith Ian Price wrote: >>>>> >>>>>>On August 22, 2002 at 10:04:37, Peter Hegger wrote: >>>>>> >>>>>>>On August 22, 2002 at 08:10:27, Matthew Hull wrote: >>>>>>> >>>>>>>>2. Hsu's creations _slaughtered_ the computer competition...ALL OF THEM! >>>>>>> >>>>>>>They did? Where are the game scores? I know that they claimed to have scored >>>>>>>around 90% against other programs during testing, but no game scores exist for >>>>>>>these games. >>>>>>>Are we to simply take their word for it that these games actually happened? >>>>>> >>>>>><snip> >>>>>> >>>>>>>Regards, >>>>>>>Peter >>>>>> >>>>>>I talked to Hsu and asked him specifically about this. He said he did not save >>>>>>the game scores. I listened to his answer while looking him in the eye and I >>>>>>believed him. What do you base your assertion that he is lying on? I am inclined >>>>>>to take a person's word without the personal interview, but in this case I had >>>>>>that added benefit. I don't really care if DB would beat today's programs or >>>>>>not, since it does not exist any longer, but I do not like people calling Hsu a >>>>>>liar with no evidence. If you have some, please post it. >>>>>> >>>>>>kp >>>>> >>>>>Proof? Evidence? I'm not the one who made the claims about my programs prowess. >>>>>But if I did, then I'd surely have game scores to back myself up. If I didn't >>>>>have the scores then I would keep my mouth shut. >>>>>As I stated in the part you snipped, time on a supercomputer doesn't come cheap. >>>>>Atleast not cheap enough to play dozens of games just for fun. >>>>>Peter >>>> >>>>They had all the time on the computer they wanted. IBM wanted to win the match. >>>>The games were not played "for fun" but just to see the progress they had made >>>>from Deep Thought, which lost to Fritz in the 1995 WCCC. They were unimportant >>>>to Hsu beyond the results, since he had no intention of publishing them, and >>>>giving Garry any insight into possible weaknesses of Deep Blue. But whether or >>>>not he was derelict in his scientific duties or not (I suspect not, since he >>>>wasn't trying to show he could beat the current commercials with the Deep Blue >>>>project, but rather beat the human world champion), that is not what I posted >>>>about. I have found that calling someone a liar with no evidence is usually the >>>>province of liars and not honorable men, since liars think everyone lies. But >>>>since you are a pawnbroker, and perhaps have people lying to you every day, I >>>>guess you are entitled to have no trust in another's word. But you say that, if >>>>you did not have the scores, you would keep your mouth shut. Why would you >>>>refrain from making that claim without proof, when you show no compunction for >>>>posting other claims, in fact somewhat libelous claims, without even having >>>>evidence? >>>> >>>>When I talked to Hsu about this, he showed real pride in the accomplishment, >>>>which doesn't square with his making a specious claim. As I recall, he never >>>>posted and boasted about this in any fora, but merely answered a question by Bob >>>>Hyatt through email about Deep Blue's performance relative to the current >>>>commercial programs. This was after the fact, and he had not kept the scores at >>>>that time either. Bob was the one who posted the claim. I'm not sure, since it >>>>has been several years, but I think Hsu said that he wished that Bob had not >>>>posted that, since he did not have the scores. He knew there had been a furor >>>>about it on CCC. >>> >>> >>> >>>I don't think he realized it until hindsight helped him. In fact, when he >>>gave me those results, I specifically asked "Is it ok to post this to the >>>chess newsgroup?" and he responded "I don't care, it is up to you. I don't >>>consider it top secret or anything..." >>> >>>I take it just like you did. He did this. He was surprised by the outcome, >>>particularly the fact that the new chess chip, even at a slow speed, was able >>>to attack the micros (which were not oriously bad at king safety back then >>>anyway) and end most games quickly. He was happy to see the new stuff working. >>>He went on and then mentioned it much later in some sort of email exchange we >>>had... >>> >>>I did the same thing with Crafty vs Cray Blitz last year. I played 10 games to >>>see how they did, Crafty got smashed, and I thought no more about it. The logs >>>were timed out and removed from the Cray, and I never saved them as I didn't >>>think it particularly interesting to anybody with Cray Blitz on permanent >>>retired status. >> >>Bob, >>I see two aspects. I agree that it's completely ok that you did it this way. >>Perhaps a different result would have inspired you to keep the records. But this >>is all speculation. Also Hsu had all the rights in the world to do what he >>wanted. But I see a second aspect, and this one is a real problem for both of >>you! The problem lies in your own description here. >>First Hsu: >>You mention king safety. Does that mean that Hsu or anyone in his team is >>capable to watch a game of chess without keeping record of the moves and then >> at the instant judging what's going on on the board? > >I can only speak for myself. Yes, I can watch a game and decide either "king >safety looks to be working well here" or "king safety sucks and I am getting >attacked without realizing it." > >I personally used to do this all the time as I worked on that particular bit >of code... > >I can watch a series of 10 games and draw conclusions without having any kind >of game scores handy. If I thought I needed to improve something, I would >certainly want the scores so I could go back to troublesome positions and test >new changes to see if they solve the problem. > >But for example, Crafty plays GM players all the time. Scrappy played a >20 game series against one particular GM a week or two back, and won every >game. And if you wanted to see the logs you would be out of luck because >I don't keep logs from won games at all, to reduce volume... I suspect that >this is exactly the thinking of Hsu... > > > >> This is by no means a trivial >>task. Even if you watch that something is happening on the king's side, does >>that mean that you also discover the main mistakes? > >Right idea, wrong context. If you see a _problem_ then you are correct. But >suppose you are perfectly satisfied with how your king safety did during a 10 >game match? What is the point to save _those_ games? I don't because otherwise >I would now be managing over a million log files, each being about a >megabyte in size. > >> And even if you are doing >>fine so far, is Hsu or et al. able to keep the moves in their head to make some >>sound conclusions afterwards? > >Again, the wrong idea. For example, I want to try a new extension idea. I >play old vs new for many games. If I don't see any problems or issues while >the games are in progress, then I will be very unlikely to save those games. If >I notice something that it appears to have overlooked, then I will definitely >save that game so that I have a sample where it failed, so that I can test a >fix for the problem. For example, if I design a coin that should come up heads >75% of the time, and I start flipping it and it _does_ average heads 75 % of >the time, I will be happy and not save the precise trial results. I'm looking >for cases where it seems to fail so I can fix them, rather than looking for >cases where it works perfectly as they don't show me much I don't already know. > >For example, it might be nice to have all the games from the first GM vs >computer event Crafty entered. It didn't lose a single game and finished above >all the GM players. But I didn't save them as that was not particularly >unexpected. If I mention that tournament today and someone says "hey, show me >the games from that event to prove you really beat all those GMs at game/30." >I would be in exactly the same position as Hsu. I didn't save them because >their "novelty content" was zero. > > > > >> Seems not to be the case because if they were >>eidetics then they could easily reproduce the whole games even today!! > >Not necessarily. Many strong chess players can "almost" replay games from >memory. But they miss a move here and there, often making a mistake where a >piece was moved to. The mistakes don't change the game because the humans >are really remembering "concepts" or "relationships among the pieces" rather >than specific moves, but it would taint the recollection. > >However, I doubt most chess players (rated 2500) remember games against >1200 players where they won quickly and easily... Because it isn't important >to them. I suspect anybody that ever beat Fischer recalls that game perfectly. > > > > > >>So, either they had the gamescores in mind _or_ they hadn't. But then they >>couldn't soundly discuss the games. King safety and stuff like this. This is a >>job for masters, you know! Of course a little superficial fuss is always >>possible even for beginners. But is this what you would believe about Hsu et >>al.? Surely _not_! So, we have a real contradiction. What is the truth? > >First, Murray Campbell _was_ (is) a chess master. So I don't see the >issue. Second, I feel perfectly capable of watching a game and saying >either "the program handled king safety issues very well" or "the program >got pawn-rolled because it didn't quite understand the g4/h4 pawns until >they advanced farther.." > >I think most chess programmers can make reasonable deductions from watching >games, otherwise they are not going to be good chess programmers... > > > > > >>Then you, Bob: >>You say you didn't keep the logs. To me you once declared that keeping the logs >>of all your games would depass the capacity of all actually available discs. But >>what is with the chess content of such games? Look, we simulate even games >>against the long-time dead Capablanca or Alekhine. And you, Bob, are no longer >>interested in some games from CRAY? > >Not when I knew that the T90 was far faster than my quad xeon. IE I don't >particularly watch my son at the drag strip when his mustang is going up against >someone's hot-rod pickup truck. I _know_ how the race will end and don't find >it particularly worth watching nor remembering. Against a Corvette or against >a camaro/firebird, it is a different issue, because the outcome is now >variable and will rely on our preparation vs our opponent's preparation. > > >> Your CRAY?? And also this - say, you >>observed a crucial tendence how exactly CRAY crushed Crafty, wouldn't it be >>interesting how exactlx it worked? This didn't interest you? I can't follow you >>_if_ all this should be true. As a chessplayer I ask you as a chessplayer, are >>you never interested in the chess? Of course you are! And exactly when you play >>CRAY (!) the chess of the games is uninteresting? >> >>Please give us a deeper view into such questions. This would be important in >>special for those here, who will never be able to work on such machines. Thanks. >> >>Rolf Tueschen >> >> >> >>> >>>I later mentioned this in responding to someone about Cray Blitz, and got a >>>lot of criticism for not keeping the logs. But at the time, they didn't seem >>>important nor useful. Crafty has played at least a Million games on the >>>chess servers. I didn't save _any_ of the old logs. Too much data. Too much >>>space. No reason to do so... But of course, one day, there will be a good >>>reason for wishing I had a few. :) >>> >>>I don't think the DB guys lie. I don't think they cheat. I think they set out >>>to do a specific task, spent all their time preparing, and finally delivered the >>>promised result in 1997. And some _still_ refuse to accept it... >>> >>> >>> >>> >>>> >>>>The reason for my snip was not to take away from your post, but because my post >>>>was an answer to the question, "Are we to simply take their word for it that >>>>these games actually happened?" My answer is 'yes'. It is something that most >>>>people do, unless they have been shown good reason not to. >>> >>> >>>Correct. There are lots of reasons for _not_ believing them. To produce >>>troll after troll, since there is no way to _prove_ that the troller is really >>>trolling. That is the most common reason. There are others that are probably >>>obvious to most. >>> >>> >>> >>>> >>>>kp
This page took 0 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.