So lately it is getting harder and harder to know one’s way around all the strong neural net based AIs that we have in the time as new ones pop out on occasion. I propose we all share information on new ones including short description and information as to where to get them. I would prefer for us not to clutter the topic with strong AIs you can’t actually download and use at the moment like fineart
Gonna start off with obvious ones.
Leela Zero - open source community driven application of AG Zero paper
Leela Master - Weights for Leela Zero that have more human style - trained with human games on top of ai self plays. As a lot of games are from tygem it gravitates towards agressive style with big semeais.
Now out of developement (probably) strong AI. Very nice and natural playstyle in opening
Leela 11 Bit behind the times at this point but still good, roughly strength of very strong amateur - uses fuseki library
Another implementation of AG Zero paper
yet another implementation of AG Zero paper - weights can be run with Leela Zero program
Don’t have the link but we can’t leave out Facebook Elf
He just did. :o As well as PhoenixGo (top-pro+ level bot developed by the Chinese megacorp Tencent) and a number of proprietary bots.
Not to be too salty about this. But how about posting description and link rather than just complaining?
I totally would, but I can’t edit your OP. It’s probably best to have a reference list in the OP, no?
I can but I would really appretiate description in two sentences maybe next time . I have no idea what pheonix go is and it will take me time to research whilst you already know it.
Thank you for adding Elf,
Not sure if it is exactly what you’re interested in, but I think it might be worthwhile mentioning Lizzie which is a very nice UI that can run any of these weights by default it ships with LeelaZero (hence the name) but you can replace the weights file with another simply enough.
Zenith Go 7 - First pro-level commercial software. CPU-accelerated neural network trained with human games. Variable board size and komi. About 9 dan.
Crazy Stone Deep Learning - CPU-accelerated neural network trained with human games. Variable board size and komi. About 7 dan.
For PheonixGo, converted weights can be run with Leela Zero.