Sign up to save your podcasts
Or
Share “Some data from LeelaPieceOdds” by Jeremy Gillen
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
“Some data from LeelaPieceOdds” by Jeremy Gillen
I've been curious about how good LeelaPieceOdds is, so I downloaded a bunch of data and graphed it.
For context, Leela is a chess bot and this version of it has been trained to play with a handicap.
This is BBNN odds, meaning Leela starts without bishops and knights.I first heard about LeelaQueenOdds from simplegeometry's comment, and I've been playing against it occasionally since then. LeelaPieceOdds is the version with more varied handicap options, which is more interesting to me because LeelaQueenOdds ~always beats me.
Playing Leela feels like playing chess while tired. I just keep blundering. It's not invincible, and when you beat it it doesn't feel like it was that difficult. But more often than not it manages to overcome the huge disadvantage it started with.
The data
Bishop-Bishop-Knight-Knight
How to read the graphs
Orange means the human player won. The title contains the handicap[1] and the time control. I split the data by time control (Bullet, Blitz, Rapid), which indicates how much time the human is allowed to spend thinking over the whole game. Leela always plays quickly so the time control is essentially only affecting the human.
For reference, these are the time control categories [...]
---
Outline:
(01:00) The data
(01:03) Bishop-Bishop-Knight-Knight
(01:43) How to read the graphs
(03:13) Knight-Knight
(03:51) Bishop-Bishop
(04:32) Rook-Knight
(05:04) Bishop-Bishop-Knight
(05:26) Rook-Knight-Knight
(05:46) Rook-Bishop-Bishop
(06:09) Queen-Knight
(07:09) Queen-Rook
(08:00) Queen-Knight-Knight
(08:21) Queen-Bishop-Bishop
(09:01) Queen-Rook-Knight
(09:41) Queen-Rook-Rook
(10:43) Thoughts
(10:46) I think the stats are biased in favour of Leela
(11:28) How much room at the top?
(12:08) Relevance to AI control cruxes
(13:17) Relevance to cruxes with people outside the alignment sphere
The original text contained 4 footnotes which were omitted from this narration.
---
First published:
October 29th, 2025
Source:
https://www.lesswrong.com/posts/eQvNBwaxyqQ5GAdyx/some-data-from-leelapieceodds
---
Narrated by TYPE III AUDIO.
---
Images from the article:
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
View all episodes
By LessWrongI've been curious about how good LeelaPieceOdds is, so I downloaded a bunch of data and graphed it.
For context, Leela is a chess bot and this version of it has been trained to play with a handicap.
This is BBNN odds, meaning Leela starts without bishops and knights.I first heard about LeelaQueenOdds from simplegeometry's comment, and I've been playing against it occasionally since then. LeelaPieceOdds is the version with more varied handicap options, which is more interesting to me because LeelaQueenOdds ~always beats me.
Playing Leela feels like playing chess while tired. I just keep blundering. It's not invincible, and when you beat it it doesn't feel like it was that difficult. But more often than not it manages to overcome the huge disadvantage it started with.
The data
Bishop-Bishop-Knight-Knight
How to read the graphs
Orange means the human player won. The title contains the handicap[1] and the time control. I split the data by time control (Bullet, Blitz, Rapid), which indicates how much time the human is allowed to spend thinking over the whole game. Leela always plays quickly so the time control is essentially only affecting the human.
For reference, these are the time control categories [...]
---
Outline:
(01:00) The data
(01:03) Bishop-Bishop-Knight-Knight
(01:43) How to read the graphs
(03:13) Knight-Knight
(03:51) Bishop-Bishop
(04:32) Rook-Knight
(05:04) Bishop-Bishop-Knight
(05:26) Rook-Knight-Knight
(05:46) Rook-Bishop-Bishop
(06:09) Queen-Knight
(07:09) Queen-Rook
(08:00) Queen-Knight-Knight
(08:21) Queen-Bishop-Bishop
(09:01) Queen-Rook-Knight
(09:41) Queen-Rook-Rook
(10:43) Thoughts
(10:46) I think the stats are biased in favour of Leela
(11:28) How much room at the top?
(12:08) Relevance to AI control cruxes
(13:17) Relevance to cruxes with people outside the alignment sphere
The original text contained 4 footnotes which were omitted from this narration.
---
First published:
October 29th, 2025
Source:
https://www.lesswrong.com/posts/eQvNBwaxyqQ5GAdyx/some-data-from-leelapieceodds
---
Narrated by TYPE III AUDIO.
---
Images from the article:
Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
More shows like LessWrong (30+ Karma)
View all
Making Sense with Sam Harris
26,330 Listeners
Conversations with Tyler
2,453 Listeners
The Peter Attia Drive
8,557 Listeners
Sean Carroll's Mindscape: Science, Society, Philosophy, Culture, Arts, and Ideas
4,182 Listeners
ManifoldOne
93 Listeners
Your Undivided Attention
1,601 Listeners
All-In with Chamath, Jason, Sacks & Friedberg
9,927 Listeners
Machine Learning Street Talk (MLST)
95 Listeners
Dwarkesh Podcast
511 Listeners
Hard Fork
5,512 Listeners
The Ezra Klein Show
15,931 Listeners
Moonshots with Peter Diamandis
545 Listeners
No Priors: Artificial Intelligence | Technology | Startups
131 Listeners
Latent Space: The AI Engineer Podcast
94 Listeners
BG2Pod with Brad Gerstner and Bill Gurley
467 Listeners