- Blind mode tutorial
lichess.org
Donate

Using Maia and Leela Chess Zero to Find Repertoire Gaps

What are the results , though, you tease!?? :D

What are the results , though, you tease!?? :D

@TotalNoob69 said in #2:

What are the results , though, you tease!?? :D

Surprisingly good and super useful, it runs a 1000 move Repertoire in around a min.

I have fully implemented this in the repertoire builder and the test users are enjoying the results and finding gaps, it will be release to everyone in January for everyone.

@TotalNoob69 said in #2: > What are the results , though, you tease!?? :D Surprisingly good and super useful, it runs a 1000 move Repertoire in around a min. I have fully implemented this in the repertoire builder and the test users are enjoying the results and finding gaps, it will be release to everyone in January for everyone.

I am finding it is working very well. The scan of gaps using Maia2 is successfully identifying probable human moves in the repertoire(s) that I have not yet prepared (which I can confirm against the databases we have available directly within rep builder). I can couple that with engine scans to identify objectively strong opponent moves that have also been neglected to try to mitigate falling victim to opponent engine prep against my lines. Really happy with this addition to rep builder. Great work @HollowLeaf

I am finding it is working very well. The scan of gaps using Maia2 is successfully identifying probable human moves in the repertoire(s) that I have not yet prepared (which I can confirm against the databases we have available directly within rep builder). I can couple that with engine scans to identify objectively strong opponent moves that have also been neglected to try to mitigate falling victim to opponent engine prep against my lines. Really happy with this addition to rep builder. Great work @HollowLeaf

@mattchessic said in #4:

I am finding it is working very well. The scan of gaps using Maia2 is successfully identifying probable human moves in the repertoire(s) that I have not yet prepared (which I can confirm against the databases we have available directly within rep builder). I can couple that with engine scans to identify objectively strong opponent moves that have also been neglected to try to mitigate falling victim to opponent engine prep against my lines. Really happy with this addition to rep builder. Great work @HollowLeaf

Yeah, I am actually surprised but how well it works, I did not expect the speed and the results. And the toggle between Maia and Lc0 works a treat.

@mattchessic said in #4: > I am finding it is working very well. The scan of gaps using Maia2 is successfully identifying probable human moves in the repertoire(s) that I have not yet prepared (which I can confirm against the databases we have available directly within rep builder). I can couple that with engine scans to identify objectively strong opponent moves that have also been neglected to try to mitigate falling victim to opponent engine prep against my lines. Really happy with this addition to rep builder. Great work @HollowLeaf Yeah, I am actually surprised but how well it works, I did not expect the speed and the results. And the toggle between Maia and Lc0 works a treat.

@HollowLeaf said in #3:

What are the results , though, you tease!?? :D

Surprisingly good and super useful, it runs a 1000 move Repertoire in around a min.

I have fully implemented this in the repertoire builder and the test users are enjoying the results and finding gaps, it will be release to everyone in January for everyone.

sounds amazing, looking forward to it!

@HollowLeaf said in #3: > > What are the results , though, you tease!?? :D > > Surprisingly good and super useful, it runs a 1000 move Repertoire in around a min. > > I have fully implemented this in the repertoire builder and the test users are enjoying the results and finding gaps, it will be release to everyone in January for everyone. sounds amazing, looking forward to it!

@Ahmed_Mansour said in #6:

sounds amazing, looking forward to it!

Yep, it is already on my test platform, it will be refined before general release in Jan.

@Ahmed_Mansour said in #6: > sounds amazing, looking forward to it! Yep, it is already on my test platform, it will be refined before general release in Jan.

This idea can actually become revolutionary in chess improvement. Most of the people struggle with finding mistakes in their gameplay, making a good analysis to find gameplay gaps. Imagine an AI coach that actually understands human moves and looks through your games like a real, human coach.

This idea can actually become revolutionary in chess improvement. Most of the people struggle with finding mistakes in their gameplay, making a good analysis to find gameplay gaps. Imagine an AI coach that actually understands human moves and looks through your games like a real, human coach.

@yonaltimur said in #9:

This idea can actually become revolutionary in chess improvement. Most of the people struggle with finding mistakes in their gameplay, making a good analysis to find gameplay gaps. Imagine an AI coach that actually understands human moves and looks through your games like a real, human coach.

But at the same time, people won't put effort into finding mistakes. They won't improve. Have you ever wondered why chess players were stronger before then now? Because they put all their effort into finding mistakes, they fought for improvement.

@yonaltimur said in #9: > This idea can actually become revolutionary in chess improvement. Most of the people struggle with finding mistakes in their gameplay, making a good analysis to find gameplay gaps. Imagine an AI coach that actually understands human moves and looks through your games like a real, human coach. But at the same time, people won't put effort into finding mistakes. They won't improve. Have you ever wondered why chess players were stronger before then now? Because they put all their effort into finding mistakes, they fought for improvement.