„A neural net chess engine that’s been specifically trained to play like a human of a given level would, by using online games of that level as the corpus“
In the past a „normal engine“ was either unbeatable or played somehow unnatural. This engine could be a gamechanger. What do you think?
And I would suggest to set slowmover UCI option to zero; another way to comply with the Github readme.md for Maia:When using the model in UCI mode add nodes 1 when querying as that disables the search.
This trick has been disabled with lc0.exe versions higher than 24.1, so better use v23 or v24.1 with Maia, should you like it to play accordingly to its strict net policy, without regards to the thinking time set in the GUI you use.
Best,
Tibono
I take it that she was made to run on cpu for those levels.
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers
„A neural net chess engine that’s been specifically trained to play like a human of a given level would, by using online games of that level as the corpus“
In the past a „normal engine“ was either unbeatable or played somehow unnatural. This engine could be a gamechanger. What do you think?
Peter
There is no any "holy grail".
If there would be an NN engine with real humane style it would be the FatFritz.
Net of Fat Fritz was trained partly or fully based on human (chessbase Database) games but FatFritz "style"
is a typical NN-style. To make an engine with human style at first it must make a human "engine" what can play chess too.
That is all.
Maia Chess. I played some games with maia9 (=1900 Elo). It plays really human-like. I can NOT see a difference. I am playing chess for 40 years now. This is the first and ONLY engine that makes inaccuracys/mistakes/oversights a human (in this certain Elo-Range) would make in reality. I am REALLY very impressed! I am very curious - what comes next?
On Twitter they wrote they are working on a bigger/better opening book based on human games in the certain Elo-Range.
PeterO wrote: ↑Tue Dec 08, 2020 2:39 pm
Maia Chess. I played some games with maia9 (=1900 Elo). It plays really human-like. I can NOT see a difference. I am playing chess for 40 years now. This is the first and ONLY engine that makes inaccuracys/mistakes/oversights a human (in this certain Elo-Range) would make in reality. I am REALLY very impressed! I am very curious - what comes next?
On Twitter they wrote they are working on a bigger/better opening book based on human games in the certain Elo-Range.
Peter
I expect somebody to come along and try to create a Magnus Carlsen clone with Maia for the PlayMagnus app.
thats for sure. I would love a Maia-Fischer - Maia-Spassky rematch!!!!
I think we will see ALL chess legens - Kasparov, Karpov, Aljechin ....
I love to play for my own - so I need some personalities with Elo 1900 - 2100. So I can play a tournament with stronger and weaker oponents (my Elo is about 2000).
PeterO wrote: ↑Tue Dec 08, 2020 7:37 pm
Hi Madeleine,
thats for sure. I would love a Maia-Fischer - Maia-Spassky rematch!!!!
I think we will see ALL chess legens - Kasparov, Karpov, Aljechin ....
I love to play for my own - so I need some personalities with Elo 1900 - 2100. So I can play a tournament with stronger and weaker oponents (my Elo is about 2000).
I am fascinated by this perspectives
Peter
Sadly, i don't think we have enough played games of each famous players to have a reliable maia net. The player with most games in chessbase megadatabase 2019 is Korchnoi with around 9000 games if I remember right. Other famous players being 5000 games and less
Hope I am wrong since I don't really know how many games are needed and if a different net layer could do the job with few games.