terry mcintyre wrote:
Thank you! At present, computer go programs may be strong relative to
each other, and they may actually beat some humans of moderate
ability, especially at timescales too quick for amateur humans, but
most programs also have high-kyu-sized gaps in their knowledge,
including seki and nakade concepts.
We won't see programs regularly beating pros until those gaps are
filled.
Maybe, but maybe it's not even needed.
Each time Leela gets online after absence on KGS, it gets flooded with
players who have observed the games and who think like you, i.e. "the
computer is only rated 2k because some amateur fools made huge mistakes
in quick games, so let me use the fact that it's horribly overrated to
improve my own rating".
The net result is usually that Leela rating goes up more because those
players end up face down in the mud. It's only when players who have
played it before face it, that the correction sets in.
I've said it before and I'll say it again: it is perfectly possible to
have huge knowledge and ability gaps and still be an extremely strong
player. There are ample examples. Computer chess programs still suck
horribly in some blocked positions.
Substituting billions of playouts for a life-or-death or seki
analysis which 10 kyu players can manage in seconds is inefficient;
computers could be doing something more effective with that time.
Due to the nature of the problem no reasonable amount of playouts helps
Leela, so I'm not sure what this is supposed to mean.
I understood for example MoGo is now better at this and still uses Monte
Carlo.
So I think the jury is not out yet whether this really needs another
approach.
--
GCP
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
