Author: Peter Kappler
Date: 09:34:44 12/07/99
Go up one level in this thread
On December 07, 1999 at 09:59:08, Charles Unruh wrote: >On December 06, 1999 at 20:46:10, Peter Kappler wrote: > >>On December 06, 1999 at 10:18:00, Charles Unruh wrote: >> >>>On December 05, 1999 at 18:44:57, James Robertson wrote: >>> >>>>On December 04, 1999 at 16:40:02, Charles Unruh wrote: >>>> >>>>>How much does it take to show blind men that Comps are GM strength. A program >>>>>beats a GM, draws several more, then beats lithuanian national team, Draws a >>>>>40/2 with Anand and there are people here who want to make out that it's hardly >>>>>USCF master strength!!! >>>> >>>>Rebel must be GM strength. How else could it lose to the Lithuanian National >>>>team, lose to Anand, lose to Rohde, lose to Hofman, and lose to who knows who >>>>else. >>> >>>Oh Kasparov lost to sokolov, that must mean he's no where near to 2830 right >>>> >>>>If we just count Rebel's victories it is unquestionable super grandmaster. If we >>>>count just its losses it is unquestionably 1500. Average them out, and you get >>>>IM. >>> >>>It would be one thing if these victories where all spaced out. However for >>>rebel to get it's wins all in a short period of time decreases the odds. If a >>>Human were to draw ANAND, beat the lithuanian nat'l team, Beat sherbakov, almost >>>beat a 2593 Baburin, all in a few months there is no way that you would be >>>trying to make out that he wasn't GM strength. >> >>Why do you continue to only focus on the wins? James is right, you must >>consider *all* results, not just the ones that support your viewpoint. >> >> >> You can not find an IM in the >>>world that could produce the same sort of results in the same period of time! >> >>Sure you could - > >No you couldn't not against similar competition, in the same amount of time. >Sure an IM could get a draw against Anand maybe 1 in a 100 games. This is completely wrong - you need to understand the ELO rating system before you make these kinds of statements. Even at a 400 point rating differential, the lower rated player has a win expectancy of .091, meaning that a mere 2370 player could expect to draw 2 games out of 10 against Anand. The >likelyhood however that the draw that rebel got out of two games was that 1 in a >100 seems a bit slim. Indeed 1/100 would be slim, but as I explained above, the odds are *much* better than this... What's even more amazing is that Just as Bertil Eklund >has said these are results in match play. Rebel would be even stronger in a >swiss system even where all these opponents had not come prepared to play it in >specific. I have it on no less an authority than Vagr (Vladamir Akopian GM) >that "Computer programs are at least low to mid-level Grandmaster strength." I'm not 100% certain that Rebel *isn't* playing at GM strength. Frankly, I think there still haven't been enough games to know for certain. Akopian's opinion is one I would value highly, but in the end, the only evidence that really matters is how Rebel performs in 40/2 games against strong humans. So far it has achieved a performance rating of 2485, and that isn't quite GM strength. (And certainly not high enough to earn a GM norm) >I'm sure that you have even seen Kaufman saying they were GM strength several >software and processor generations ago. Even Robert HYATT back when we had a >mere Hiarcs 6 running on a p200 said then he thought they were 2450(AND IT IS IN >THE RECORD!). Sure, fine - that's Bob's opinion. By the way, I think Bob tends to quote these numbers as USCF ratings, so you'd need to deduct ~75 points to translate that to a FIDE rating. Again, I don't care about people's opinions. We have lots of real results now, so lets focus on them. Though strangely, even though, We have 1 gigahertz machines >commercially available, and 2 software generations later, he and people like >you don't want to admit that they've improved one point. Comps are GM's have a >coke and a smile. > Maybe. The difference seems to be that you are willing to accept this on faith, while I'd like to see more evidence... >in fact, I think somebody recently posted that Rebel's >>performance rating in the recent 40/2 games is around 2485. This sounds like a >>solid IM level performance to me. >> >>>Do you know Anand has not drawn a player that was only IM strength in years! >> >>Ha! Maybe because he almost never plays IMs! >He has played several though, and i wonder why they didn't just happen to get >that lucky one in a hundred draw hmmm? > You have this tendency to throw around "facts" rather loosely (see the 1/100 silliness above). Can you please post a list of the IMs that he has played recently? If you can't post it, then please stop making this argument. My guess is you'll only be able to find a very tiny sample of games, in which case the result is particularly meaningful. >Super-GMs tend to make 3 or 4 >>tournament appearances a year, usually in Category 18-20 events where they face >>nothing but other super-GMs. So this "statistic" of yours is of questionable >>value. >> >> >>>Further you can't totally judge a program by results. >> >>You might want to reconsider this statement... >> > >You say I might reconsider and half a paragraph down admit i'm right, well thank >you i appreciate that. > Wrong again - my comment was that it was "basically a non-issue" against GM-strength players. >> >> Why you ask? Becasue the >>>comp doesn't know who you are. An example of what i mean is this I have had >>>several draws against Comps by 3 repetition. The comp takes the draw because >>>the position is basically equal and it doesn't KNOW that i'm not a GM. A GM >>>would break the rep, even if it was a slightly inferior move because he knows >>>i'm only 2000 and he can beat me. It's not that the comp wouldn't destroy me if >>>played an alternative move to the rep, yet this would lower the rating. >> >>This is an easy problem to fix - > >Oh so it's a problem like i said, imagine that. > >most programs let you set a "contempt factor" >>which will tell a program to avoid a draw unless the score drops below a certain >>value. I might add that while this might be an issue in games against weaker >>players, it's basically a non-issue against GMs. >> > >It's a total issue, with IMs, GMs, and everyday patzers. This sort of thing >doesn't just happen with perpetual check. It happens In many positions because >the computer chooses moves that it supposes are the best instead of a move that >might be objectively inferior that leads to complications to beat a weaker >opponent because it doesn't know that you are not Kasparov. We will have to agree to disagree on this one, because it's purely a matter of opinion. My position is that the "contempt" problem you described above is no different than a dozen other quirks of chess computers. You might as well be arguing that they aren't good at long-range positional planning. A computer is what it is. --Peter >> >>>> >>>>This is true for ANYBODY. Don't blindly filter out the results you don't want to >>>>see and claim that you have discovered something amazing. >>> >>> >>>I wish you would stop blindly filtering out results you don't want to see and as >>>for this line "and claim that you have discovered something amazing." I would >>>imagine that their are a good number of psychiatrist in your local area you >>>might want to meet with to discuss your ongoing confusion. >> >>Why do you have to start insulting him, Charles? He made a perfectly valid >>objection to your line of reasoning. >> >>--Peter > > >His claim is bogus, and this statement "and claim that you have discovered >something amazing." is moronic, and at the same time attempting to be >smartassholic(word construct). So if you spit fire you get fire.
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.