Author: Mark Young
Date: 06:53:42 07/26/98
Go up one level in this thread
On July 26, 1998 at 09:18:49, Robert Hyatt wrote: >On July 26, 1998 at 08:28:08, Shaun Graham wrote: > >>On July 25, 1998 at 19:55:45, Robert Hyatt wrote: >> >>>On July 25, 1998 at 11:04:40, Shaun Graham wrote: >>> >>>>"At 40/2 they are not GM yet, but they are pretty close, and if the human GM >>>>doesn't take care, he can get rolled up pretty badly, since the computer is >>>>going to be quite attentive toward the least tactical mistake, where another >>>>human might miss it entirely. The better they (the programs) get, the harder >>>>it will be to attract human GM players to play them." >>>> >>>>(A quote from Robert Hyatt) >>>> >>>> He says "pretty close" now, that's not what my memory recalls him saying. >>> >>>Depends on your definition of "pretty close". I have said "FIDE 2400" for quite >>>a while now. Which is still a ways from the minimum 2500 needed for a GM title. >> >> " In the 2 slow games Anand didn t show any overwhelming >>superiority, as he would have against a 2400 player. I have no doubt about Anand >>being stronger than Rebel 10, but not by 400 Elo points. During 1997 and 1998, >>Anand drew 48 games playing white. The lowest ranked opponent in these games was >>Ljubojevic with 2565. Then Piket with 2575 and Hübner with 2580. All others were >>rated 2630 or higher. Of course, all well known GMs. I don t say this as proof >>of Rebel 10 being a GM, but overall as strong indications that its real strength >>is well above 2500." (Enrique Irazoqui) >> > > >for a rating, I discount "first games". I've taken too many "first versions" >into competition and saw them do very well, only to see the humans "catch on" >after a few games and turn things around. IE you can take Crafty, put it on a >server, and change anything of your choice and its rating will almost instantly >go up. But it will likely drop later. I once screwed up king safety, and when >I logged on, ICC was abuzz with "wow, watch this thing, it has busted 3 GM's in >4 game matches, and it is attacking like mad." > >It was. But it wasn't long before it became apparent that its attacks were >generally unsound, and the GM's picked this up (after you get a g4/h4 shoved >in your face by a program that is a deadly calculator, you can get intimidated) >they began to smash it, until I found and fixed the bug. Ditto for opening >book selection. Everyone has found that a new book produces a jump in rating >until players "figure it out." > >With that said, I personally will wait until Rebel 10 is released, and it has >the opportunity to play several GM players on the chess servers. Rebel 8 had >serious king-safety problems against IM/GM players. I haven't seen any Rebel 9 >users playing on ICC so I have no opinion there as of yet. But if there is a >hole in Rebel 10, it will become apparent after a few games against several GM >players. Then we can figure out if it is a 2300, 2400, or 2500 player. Note >that Crafty has absolutely crushed GM players even in game/30, yet *I* don't >believe it is a 2500 player, unless you restrict it to game/30 and faster. But >at 40/2hr, things are different... > > > > >> >>> >>> >>> >>>>Regardless of that however, what does it mean? "Pretty close" to a Shirov? >>>>Certainly not. "Pretty close" to a Kaidanov or Gulko? Hmm almost certainly >>>>not. "Pretty close" to a Kempinsky, Groszpeter, or Morovic(GMs you have >>>>probably never heard of)? Well the truth is that these latter GM's would have >>>>(probably) been toasted by Rebel 10 if they had played it 2 40/2 games. >>>>Regardless of what statistics say how often would you think Anand fails to beat >>>>2500 rated GMs? And i do mean beat them handily, not a situation where everyone >>>>is wondering who is winning as occurred during the Anand Rebel 40/2 games. Now >>>>of course the draw that Rebel got could have been luck, it could have even been >>>>the 1 out of however many games a "maybe weak IM"(Robert Hyatt, 1998) might have >>>>been statistically expected to draw in a match with a GM of Anand's caliber(more >>>>games are certainly needed to be definitive). Anands caliber bieng World >>>>Champion caliber. To illustrate what i mean by this(World Champion Caliber) i >>>>will quote Kasparov reffering to another GM. >>>> >>>>"I had a big discussion with my seconds over lunch about whether to play my new >>>>plan against Shaked. I would have preferred to see another player's face across >>>>the board after 13...Rd8--not necessarily Karpov, >>>>but ANY STRONG PLAYER. IT WAS lIKE USING AN ATOM BOMB TO SHOOT BIRDS."(Inside >>>>chess magazine) >>>> >>>> The so called bird, that Kasparov is reffering to is none other than the >>>>current WORLD junior champion GM Tal Shaked. Perhaps Kasparov is using a bit >>>>of bravado(??). Hmm nope Shaked stood no chance whatsoever. Yet we have just >>>>been witness to a match where a program (Rebel 10), first drew a game, and then >>>>put up an amazingly staunch resistance, so staunch in fact that Anand famous for >>>>his speed used as much time as his computer opponent. When we see such a >>>>performance against a player of "WORLD CHAMPION CALIBER" by said program we can >>>>definitely feel safe in positing the likelyhood that programs such as >>>>rebel10/Fritz5 are indeed GM strength. Especially when we can feel certain that >>>>if we took the weakest GM and paired him against the mighty Anand the outcome of >>>>the match would have indeed in all likelyhood been far far more clear. >>> >>> >>>You can stick with your opinion, of course. And I will stick with mine. I >>>simply see too many holes at present, in the micros. They have their moments, >>>and Rebel certainly played well. >> >>It is best not to (simply) attempt to stick with ones oppinions but rather to >>follow the very important concept called "the weight of the evidence". Based >>on two games against Anand the weight of the evidence is not overwhelmingly >>heavy in either direction. However, from these games one would be hard pressed >>to form a hypothesis that "the likelyhood is Rebel10 is not GM strength". >>Indeed the evidence would lend to one forming the exact opposite hypothesis. >> > > >As I said, I have done this. Rebel 10 doesn't offer us enough evidence yet. >Rebel 8 was nowhere near a GM level. I've seen nothing that says Rebel 9 is >a quantum leap. Ditto for Rebel 10. Steady improvement? Probably. But until >I see it play 25-30 games against 2500 players, and roughly "break even" with >them (or better) I'll withhold judgement. The only two games I have so far >are a loss and a draw against Anand. That's not enough information for me. >Yet... > > > >> >> But I'd still bet on Anand, after giving him >>>a few games to see how it plays.... >> >>As i look in this post i see no hint or even dream of a suggestion that Anand is >>not clearly superior to Rebel10. In fact if it isn't as strong as a kaidanov or >>Gulko as mentioned previous i think Anand bieng stronger can be taken for >>granted. >> >> >>> >>>Unless you talk about fast games. I just did some history tests on ICC and >>>found that, for example, that Crafty is winning 3 of every 4 games from GM >>>Dlugy... that means Dlugy+200 for a rating estimate. But that is blitz. I >>>have similar results against Yasser, Roman, etc... >> >>I see no mention of anything other than 40/2 games in this post. Also in other >>post in this thread even more 40/2 game performances are attested to by other >>CCC members > > >If you only look at 40/2, what conclusion do you draw from one loss and one >draw? Performance rating? roughly 2600 after two games. But with just two >games I don't feel comfortable speculating about a program's rating. It will >be possible after it plays some where we can see more games... That's one >reason the Fredkin prize required 25 games for the >2500 rating. A couple of >games could produce any rating. And if you play 10 computers against Anand, >the odds are one would win both games on occasion. and have a 3200 rating for >two games... > > >>> >>>But 40/2 is something else... and I watched an IM (A pretty good one) rip >>>Rebel 9 badly in two games today. I don't know what kind of hardware, however, >>>as the rebel user was logged on as a guest. But this particular IM is quite >>>good against computers... I have been playing another IM some long games on >>>ICC and am breaking even pretty much, although I (Crafty) can totally shred him >>>at 5 3 and so forth... >> >>Why are you talking about how they are doing at faster controls there is even >>more evidence that they are GM's at this speed, Anand was trashed in the fast >>games, and the above game you mentioned by the IM in all likelyhood was but >>another one of these quick games. As for crafty playing an IM at long games, >>Crafty i'm sure is a competent opponent (the latest crafty anyway (it's stronger >>IMHO). I saw crafty 14.? lose a match 6 0 at 40/2 against chessmaster). >>However Crafty isn't in the same league as Rebel10 or Ftitz5. > > >Here's my point: Computers are clearly stronger as the time control gets >shorter. If a computer has trouble with an IM at roughly game/30, then it >is *certainly* going to have trouble with that same IM at game/2hr. > >And given my current hardware advantage, since I do a parallel search and >Rebel/Fritz don't, I'd take your match request whenever you want to try it. I would like to play a match with Crafty at 40 moves in 2 hours. Not out of disrespect, but respect. I would love to see how it does at slow time controls vs say Fritz 5 running on a P II 266 or P II 300. If you want to do it I'm game. >I can quite easily get 2-3M nodes per second on available hardware. That's >a big advantage. You don't think speed is important, based on Comments by >Ed in years past? Ponder this then: "why do you suppose he went to Kryotech >to get a souped-up AMD machine?" Maybe speed *does* matter after all? > >BTW, I've never seen crafty lose a match 6 0 to Chessmaster, although with 6 >games it is possible. I've also seen it win such matches regularly. CM5000 >is a good program, as good as rebel and the others, contrary to popular opinion. >Check out the current Korrespondence Kup. And look to see who's winning and who >is beating who... I wouldn't discount *any* program so quickly...
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.