Author: Dann Corbit
Date: 22:39:58 08/30/99
Go up one level in this thread
On August 31, 1999 at 01:26:40, Michel Langeveld wrote: >On August 30, 1999 at 22:28:59, Dann Corbit wrote: >>There are now 854,918 EPD positions calculated at 12 minutes of PII 300 MHz CPU >>time or more. >>Ticking down to one million positions at long time controls: >>1e6-854918 = 145,082 rows to go. > >Great!! > >Dann can't we help Kasparov-World project by evaluation a lot of positions VERY >deep with the latest crafty 16.17 and other engines. Maybe a few thousands of >position is already enough. > >I think if we have luck you can make yourself immortal by finding the winning >line. I think such a task is possible with existing tools, but not used in the pedagogic manner. In the early endgame at long time controls, computers are simply no match for a super GM. I think that from here forward computer testimony should simply be ignored unless a tactical line is seen that GM's can't refute. Here is how to beat a super GM with existing computer technology: 0. Write a computer based system that has a pool of tasks (let's say 100K tasks) 1. Computers request tasks and are assigned positions. Let's say, initially, we assign the current position and the next 3 ply. Then, at 12 minutes, we "check in" with the task assigner. We reconsider the top 3 results given all the current data. Given these three, we drop all work on all positions that are not favorable, and reassign tasks. We can work forward much deeper using a system like this than with current techniques, I think. We can get much deeper using a technique like this, but it would require massive synchronization over the net or a super computer like those made by Intel for the US government. IMO-YMMV. Of course, I am no expert at such things. This is just the mental image that I see as better than anything else we have right now. Deep Blue, fitted with 1000 of Hsu's new chips would be an excellent alternative.
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.