Author: Peter McKenzie
Date: 00:47:28 01/12/00
Go up one level in this thread
On January 11, 2000 at 12:43:13, Bruce Moreland wrote:
>On January 11, 2000 at 06:58:28, Robert Hyatt wrote:
>
>>On January 11, 2000 at 05:25:49, Amir Ban wrote:
>>
>>>On January 11, 2000 at 01:10:56, Dann Corbit wrote:
>>>
>>>>Worst of all, I think he considered Deep Blue to be a micro program with a big
>>>>box around it. He really had no comprehension of how much better 200 Million
>>>>NPS is than 200 Thousand NPS. Statments about how {paraphrasing} "computers
>>>>will never make a move like that" indicate to me that he prepared by playing
>>>>against micros. That is like preparing for Linares by running through a bunch
>>>>of games with C club players. Deep Blue will see things that other computers
>>>>simply will not see without allowing absurd time intervals. If he allowed the
>>>>micros to think for one week per move he might get something commensurate. But
>>>>then, the playing experience would not be the same, because he would be
>>>>operating at a slow, postal rate and have plenty of time to think through the
>>>>possibilities.
>>>>
>>>
>>>This has been said countless times before, and the follow up question: Show us
>>>one move that Deep Blue (or Deep Thought) made that a micro needs a week to
>>>find, has never been answered.
>>>
>>>Amir
>>
>>I pointed out one in the first match, in the game DB won, where Kasparov had a
>>mate in 1 for about 10 moves. A subtle rook move made the entire variation
>>work, where the rook move preferred by the micros at the time would have
>>resulted in deep trouble. I don't recall the game now, but I remember that
>>DB was white (again, in match 1 which it lost) and its king was hemmed in on
>>the kingside with Kasparov threatening mate. But he never got to play the
>>mate...
>>
>>Other examples are the Nolot positions. Micros get 1-2-3 maybe. Deep Thought
>>did better. And that wasn't deep blue.
>
>You'd need to document the first paragraph of this, but I agree that the Nolot
>test does show the tactical power of DT. I ran some of those positions for
>several days and didn't find solutions that DT found much more quickly.
>
>This is not true of all of the Nolot positions, however, but then again DT isn't
>*that* fast.
>
>Let's have another thread about the Nolot positions.
>
>Here are my results as of now:
Last time I ran this was about 9 months ago on a 500mhz machine using 1hr per
position, results added below:
>
> DT Ferret Chop
> 4x450 500
> ------ ------ -------
>1 6hr
>2 2min 31sec
>3
>4 2.5hr
>5 2hr
>6
>7 6hr
>8
>9 9min
>10 2min 11sec 50min
>11 5min 48sec 12min
>
>I ran mine for 3 minutes per position. As of this moment I am starting a run at
>two hours per position. Another version got position one in 1hr23min, we'll see
>how well this one does. I'll fill the chart in tomorrow. If either of you
>would like to run any of these, go for it.
I'll do another run of nolot this weekend to see how the current version does.
>
>Scoring is done for "find and hold", meaning your time stars from the initial
>fail-high, and you get no credit if you switch to some other move before the end
>of the test.
>
>bruce
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.