search

featurecat / lizzie

Lizzie - Leela Zero Interface
GNU General Public License v3.0
967 stars 228 forks source link

Winrate graph observed in the wild #142

Closed bittsitt closed 6 years ago

bittsitt commented 6 years ago

Hanayeol: "Look at the graph! Look at the graph! Ke Jie is in trouble!!" 😁 Looking good folks!

image

(Apologies for creating a new issue, but I thought this was fun :-) )

bochen2027 commented 6 years ago

Lizzie is climbing fast. I warned Sabaki devs about it but they brushed it off. they were complacent and slow to innovate.

By year end, no one on leela will be using sabaki anymore

featurecat commented 6 years ago

hahahaha!! thanks for sharing! I'm always so happy to see lizzie out and about, doing things in the wild❤️

dfannius commented 6 years ago

There is basically one Sabaki dev, and he should feel free to do whatever he wants. If anyone really disagrees with his development decisions they can fork it.

featurecat commented 6 years ago

to be explicit about it, I'm not interested in competing with or replacing other software. I really like Sabaki, I'm just trying to develop a tool (Lizzie) that meets my own wishes for an ai-based go program.

bochen2027 commented 6 years ago

I want my $80 back!

https://www.twitch.tv/videos/255566462

image

https://www.twitch.tv/videos/255566462

featurecat commented 6 years ago

haha same... I also bought Crazystone. but I'm happy to support Remi (it's developer). it was a fun purchase :) besides, looks like his latest AI might be better than Leela zero.

bochen2027 commented 6 years ago

Yes "weights" is stronger than LZ, by several hundred elo. Here is my thoughts, since LZ is already a top pro, even if his next CS was way stronger it is no longer enough to bridge the gap and still be commerically successful when there are already free top pro level bots out there (LZ etc).

During 2013 to 2016 CS occupied a unique niche in the Go marketplace in that at the time it was state of the art publicly available software and there was no equal to complete with it in that class. That was its entire value proposition. But now all that has changed. Even if Weights is stil much stronger than LZ, what is the time to market? by the time the new CS comes out, maybe LZ would have already trumped it by then

And traditionally CS was never known for its UI either, remind me of windows 98 interface and no dpi scaling means horrible on 4k monitors etc. So he would have to improve the UI as well, otherwise its no show stopper out the door in the day and age when we have things like open source sabaki/lizzie etc... esp if it still has DRM and locked down without gtp support. (assuming CS is even going to go commerical again, which I personally don't think so, that era has passed)

BTW when are you going to refactor and then switch over to electron and dump the java? Java is like flash, both are dying breeds

featurecat commented 6 years ago

we're sticking with Java for the foreseeable future. since this project is open source, anyone may port it to whatever they'd like. I would fully support anything that was truly better :)

bochen2027 commented 6 years ago

image

CS is a good deal consider this is still selling for almost 90 bucks....

Im thinking about purchasing a license just so I can do a match aginst LZ after all it claims to be "one of the strongest" in the world......

featurecat commented 6 years ago

I think TMFOG is more about the interface than it's AI. many users like it. anyway this is not constructive.

bochen2027 commented 6 years ago

sorry about that, I was inspired by this chess video and wanted to do nostalgic comparisons

https://www.youtube.com/watch?time_continue=1&v=R3By_rdwxSg