Author: Robert Hyatt
Date: 20:33:01 01/06/00
Go up one level in this thread
On January 06, 2000 at 19:16:28, Graham Laight wrote: >On January 06, 2000 at 17:07:01, Robert Hyatt wrote: > >>>>Which perspective seems most accurate? The user of a black box, or the person >>>>that 'filled' the black box? >>> >>>Or the impartial evaluator of the black box? >> >> >>That is the point. You can _not_ evaluate the black box. You can only evaluate >>the results. The brain surgery worked. You consider it wonderful. Only the >>doctor knows all the difficulties he had during the surgery, how close he came >>to losing the patient, etc. Because the doctor sees _inside_ the black box. > >Put the human chess player is also a black box. More so, in fact! > >>That is why 'impartial evaluation' is not easy until we simply have a lot of GM >>games to go on. At present we don't. My view from inside the black box shows > >Agreed. > >>thousands of problem areas that need work. It may be that my view is wrong, if >>and only if the black box can produce results against GM players that I don't >>expect. The easy way out of this is to wait. We are getting data. We know for >>sure that Rebel isn't going to have a 2700 TPR based on games so far, so the >>2700 number for Tiger on the SSDF is grossly overinflated. As Ed said, and as I > >Ahem... aren't Century and Tiger two different programs? yes... but if you watch SSDF ratings, you will notice that a year's worth of programs are very close on the rating list... then the next year the bar goes up a notch, but all are very close overall, year-by-year. I assume Rebel won't finish far behind (or ahead of) Tiger if it gets rated by SSDF. > >The last version of Rebel to get an SSDF rating was V9. From memory, it's rating >was a little over 2500. Take away the 20-30 points for alignment with FIDE >ratings, add them back to allow for the fact that computers are faster now, and >the rating isn't far off where Ed claims Century now is. > >Having said that, I strongly suspect that Century on 400 Mhz would score >significantly higher than V9 - and I hope that the SSDF go ahead and rate it. I am sure it would score better against other computers. I am not sure it would score better against GM opponents. That is the point of this debate, of course... It is easy to play version N vs N+1 and see improvement. But when you play both against strong humans, the difference disappears most of the time.. > >>have said many times, I would consider a TPR of 2500 a remarkable result. And >>that isn't good enough to make a GM. > >True - 2550 FIDE Elo would be required. Actually you have to have a rating of 2500 during the period of time where you produce three 2600 TPR norms I believe... > >-g
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.