Thanks, Dan.
So you are running these bots yourself and getting 4-5k
rating? That surprises me a little.
Yes, I run them on a single Google Compute Engine instance (e2-micro, 2 vCPU 1 shared core 1GB memory) via just `gnugo --mode gtp`. They've both played hundreds of games by the looks of it, one is consistently strong 5k, the other very weak 4k -- I may have initially used different settings for the two but their performance against human players is very consistent to achieve these ranks.
I agree with Gunnar that a better the way to weaken GNU Go
would be to introduce randomness to the play than the
existing system of levels that reduce the processing.
(I would be more focussed on making GNU Go stronger
but that might involve rewriting it a lot.)
Yes, I am certainly not suggesting it's a good use of _anyone's_ time to make it weaker :) But I have enjoyed its style of play greatly and would love to help in any way I can. Is there any running list of bugs or features that current development would entail? Like a view similar to Issues on GitHub? I'm new to the GNU community.