Author: Robert Hyatt
Date: 12:00:53 11/12/98
Go up one level in this thread
On November 12, 1998 at 11:00:32, Amir Ban wrote: >On November 12, 1998 at 08:18:08, Robert Hyatt wrote: > >>On November 12, 1998 at 03:42:58, Amir Ban wrote: >> >>>On November 11, 1998 at 19:55:02, Robert Hyatt wrote: >>> >>>>On November 11, 1998 at 07:09:43, Amir Ban wrote: >>>> >>>>>On November 10, 1998 at 08:16:23, Robert Hyatt wrote: >>>>> >>> >>>>>> >>>>>>this is dead wrong. It overlooked a draw in game two that *kasparov* also >>>>>>overlooked. But when it played Be4 rather than Qb6, the move that Kasparov >>>>>>insisted won a pawn, it turns out that DB had seen a *very* deep draw there, >>>>>>one that Kasparov also overlooked. >>>>>> >>>>> >>>>>Don't know where this interesting but false piece of information comes from. >>>>>Deep Blue evaluated 37.Qb6 as +32 so it didn't see a draw, if there is indeed >>>>>one to see here. >>>>> >>>>>Amir >>>> >>>>Is this not the move where it chose to go into "panic time" because the eval >>>>kept dropping iteration by iteration, and it changed to Be4 at the last minute? >>>> >>> >>>That was move 36 (axb5). >>> >>>I posted the printout information here once. It contains answers to all these >>>questions. I assume you didn't save it. Since you comment on this matter often, >>>and with an air of authority, aren't you at all interested in the only data ever >>>published on this ? >>> >>> >>>>There is definitely a draw to be found... just let Junior search... for a day or >>>>so. You'll get 0.00 eventually. I did... And I assume they did as well to >>>>change to something else... >>> >>>Don't know, but if DB didn't see a draw, what does this matter ? >>> >>>I have an email from F.Friedel from May 97 saying he ran this on Fritz 4.01 to >>>ply 17 and got +0.28. >>> >>>Amir >> >> >>You keep saying that... so I'm going to respond in terms you can both understand >>*and* remember so we don't have to do this again... >> >>DB was searching Qb6... the score kept dropping. Until the last iteration >>where it changed to Be4. Now how could it do that without saying anything until >>the "reconstructing..." output? Exactly like I explained it the last time you >>brought this implied accusation up... like this: >> >>In early versions of Cray Blitz, when I started a new iteration, and the >>previous best move failed low (dropped to 0.00) I simply kept searching... and >>with no "warning" it would display and play a completely different move, because >>the fail low would put it into "extended-time mode" and it would try root move >>2, then 3 until either it tried them all and then dropped the lower bound and >>restarted the search (telling me when it did this) or it would find a new best >>move, and it might take an extended time to find this... so all I would see >>is "time limit extended because of fail low at root" (I added this so I knew >>what was going on)... followed by a brand new move out of the blue, when it >>finally found a move that produced a score above alpha. >> >>Now for the life of me I don't see why you keep making ths implication that >>something went wrong... I've pointed out several times that this is not an >>unusual thing with any program. Many of us make our output more informative >>but there is *no* mandated requirement to do so. Their output has *always* >>been confusing to me, from the * for captures, to the long-form Be4*d5 type >>of move, etc... But I certainly don't see how we go from there to "something >>odd happened that they *must* explain to prove it wasn't human intervention." >> > > >What the hell are you talking about here ? > >Who's been talking about this subject at all ? You asked if this is the >intervention "move", I answered no, that was the previous move 36.axb5, not >37.Be4. > >Let me guess: > >Did you scan my post for two seconds, make a wrong guess about what I'm talking >about, then raced your fingers to produce 6 paragraphs of a convincing and >eloquent answer to what you thought I wrote ? You must be in your "write-only" >mode. > Nope... but I did make the mistake... I had been talking about Qb6 vs Be4, which was *the* bone of contention for game two that Kasparov pointed out. Ie the move he said only Kramnik and a couple of other top GM players would ever consider or play. I overlooked your axb move... sorry... >Wouldn't you like to go back to my post and see what I really wrote ? > > >>That's baloney. It's always been baloney. It still is baloney. It will always >>be baloney... >> > >How refined > > >>It's perfectly understandable... it is only mysterious if you want it to be so. > >In your haste you didn't answer a question that I asked in that post, which I >really appreciate an answer to. So I repeat it here: > >I posted the printout information here once. It contains answers to all these >questions. I assume you didn't save it. Since you comment on this matter often, >and with an air of authority, aren't you at all interested in the only data ever >published on this ? > >Amir I saw the output a dozen times. As I said then, I didn't see a single thing that looked odd to me, based on what I had seen of their output for 10+ years. I believe I explained how they could change from one move to another "mysteriously" as you put it, without there being *any* need to suggest that something else went on inside the program. At present, about 1/2 of the programs I have seen could mysteriously change their mind with *no* warning of any kind. I've participated in enough computer events that I specifically got tired of that and now let the operator know of any fail high, or any fail low, so that he has some idea of how long he has to wait, and what is going to be played. Now, do we talk about mysteriously changing it's mind? Do we talk about the move of the game (Be4 vs Qb6) or about axb5? Either is ok. Kasparov seemed to be most concerned about move 37. That's where I spent my analysis time after that game... searching both moves to see what was happening. I didn't, and don't see anything "suspicious" unless it were my program playing the game, because in Crafty I would (a) never find that anyway in real time, and (b) I immediately would notify the operator that Qb6 had failed low at the root and that I was going to search longer to find an alternative... and I would have also reported the fail high at the root when Be4 popped up... But that is only formatting. Sorry about the move 36 vs 37 confusion... that was my fault. But the fact remains that nothing odd is to be found in the output... As I said, their output was cryptic from the first time I saw the machine play in 1987. But it was clear, even then, that they would be "something special" in the computer chess world. And they *certainly* don't deserve this continual attacking. And Shay's article was nothing but a serious insult served up to them. BTW I showed that to an economics prof here. He had some unprintable comments to make, but the gist of it was along the lines that Bruce presented. Would make a good story for the "National Inquirer" but wouldn't cut the mustard for any technical forum. But it was the last paragraph I took issue with because it was simply wrong. This is not something to get angry about. It is not something to jump up and down about. But they don't post here, and I'm going to do my best to not let such misleading stuff fall on the road and be stepped in by everyone, taking it as though it were accurate. I don't know whether I have a beef with you or not.. depends on whether you support that "article" or not. But it is not a personal beef... just a beef with something written inaccurately and without any checks on the data (the stock market comment) and with something written solely to discredit and embarass (the part about avoiding other computer events when there were none to avoid.) That's very low-quality writing. Now, back to the main deep-blue discussion?
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.