question to Prof Hyatt is Mr kaufman's method a reliable one

[duncan]

to see if rybka can come up with better positions than a gm


http://rybkaforum.net/cgi-bin/rybkaforu ... ;pid=98159

..
I think that there is a fairly good way to test the hypothesis that doesn't require a new interface or engine. First, we make a version of Rybka with only tiny eval weights, just enough to produce intelligent move ordering. This is something I can do, Vas only needs to compile it. Then we set the interface option that selects all moves within half a pawn of the best move. Since all the eval weights are tiny, this is the same as a pure list of the tactically best moves according to Rybka 3. The window could also be set to 1.5 pawns, to include pawn sacrifices, if the GM so requests. An independent person would operate this machine, and after each move by the real Rybka 3 this operator would either call out or write down all the qualifying moves after the "stupid" Rybka has some time to think (I would suggest that half of the time of the real Rybka 3 would be reasonable, as the "stupid" Rybka should analyze much faster due to more cut-offs). If the window was 0.5 pawns he would just get the moves, no eval. If he chose the 1.5 pawn window he would get the moves that were tied for best and separately those that were pawn sacrifices. He would be free to think about the game during this interval. Only after he got the qualifying move list would his clock be started. This removes any need for anyone to write a new interface or any need to discuss what program to use as the helper. The human in effect gets 50% more thinking time (although the extra time is unassisted time), which should more than compensate for any distraction involved with the process. So in summary I think this is a very fair way to test the hypothesis; all we would need is an agreement that the loser pays the GM fee, and we need an operator that both parties would accept as both competent and unbiased, and a way to compensate him unless he is a volunteer.