While this still seems far away with the state of KataGo type bots, GnuGo has a really good knowledge based engine that combines all sorts of knowledge of a move. It knows for example if a move saves a dragon (string of stones), which in turn keeps a territory of a certain size under (likely) your control.
If GnuGo had an output / explanations module, it would be a decent teacher I think, but only up to 15kyu. GnuGo is open source so this could still be made.
Thanks for the screenshots btw, seems like a pretty cool program.
I for one cannot judge though what it’s “worth” today, after all it is ~30 years old, I think, way before Go AIs changed everything. But I guess it can still teach quite well to Kyus, at least to DDKs.
Yeah, I’m honestly surprised none of the people trying to build “I hooked an LLM up to KataGo” tooling have bothered to try to work with GnuGo. It doesn’t explicitly have an explanations module, but its internal state is well-documented and it would be a neat experiment. Most of the people building these tools are usually explicitly or implicitly targeting DDK anyway, so GnuGo’s strength is probably fine.
Yes, my Go Dojo software is still being sold by me. I know the graphics are prehistoric, but the cost of upgrading it is too high relative to the sales. But questions about grouse meaning seem to have been well answered.
