AI to explain chess positions to amateurs
Moderators: hgm, Rebel, chrisw
-
JVMerlino
- Posts: 1367
- Joined: Wed Mar 08, 2006 10:15 pm
- Location: San Francisco, California
Re: AI to explain chess positions to amateurs
The Chessmaster series did it 25 years ago without using AI.
-
chesskobra
- Posts: 235
- Joined: Thu Jul 21, 2022 12:30 am
- Full name: Chesskobra
Re: AI to explain chess positions to amateurs
OK, no more sarcasm or cynicism. What OP wants is a slightly more specialised LLM. By specialised I mean something that will take Stockfish like multi-pv analysis, use game databases, and surround it with words. I think technically we are not at all far away, and is probably not difficult technically, but may need investment. Big companies like OpenAI are going after bigger markets. They must be interested in bigger science and engineering problems - drug discovery, programming, fluid dynamics, and so on. It is of course not ruled out that they will develop this; but it will be only to show off. Maybe a couple of Deep Mind or OpenAI staff can do that in 2-3 months, given their infrastructure.
Ultimately you cannot explain a position with only high level qualitative analysis; you need concrete calculation. High level analysis can tell you what a master is considering - preventing this or that move, gaining or losing a tempo, and so on. I had thought of the following idea. At the minimum what you can do is take an engine that uses hand crafted evaluation (e.g., SF 16 has the option), and say what one move gains that other doesn't, etc. This is probably doable without AI. On top of that LLM will only add some sugar. For example, you tell the AI that you want to talk about the following points 1. a3 prevents ... 2. it loses tempo ... and then the LLM will put it in a nice paragraph, add some decorative detail like it was Fischer who employed this idea first, and so on.
@chessx No more cynicism, but, whether mission impossible or not, you have to do a lot of ground work. For example, it seems you had not heard about ChessX GUI/Database and Lucas Chess. For that you have to be more broadly involved with chess. You have to also try out the various services and products to determine what they are lacking, before you make a pitch to VCs to secure funding. VCs considering funding would know a lot more about chess, game industry, and apps industry in general.
Ultimately you cannot explain a position with only high level qualitative analysis; you need concrete calculation. High level analysis can tell you what a master is considering - preventing this or that move, gaining or losing a tempo, and so on. I had thought of the following idea. At the minimum what you can do is take an engine that uses hand crafted evaluation (e.g., SF 16 has the option), and say what one move gains that other doesn't, etc. This is probably doable without AI. On top of that LLM will only add some sugar. For example, you tell the AI that you want to talk about the following points 1. a3 prevents ... 2. it loses tempo ... and then the LLM will put it in a nice paragraph, add some decorative detail like it was Fischer who employed this idea first, and so on.
@chessx No more cynicism, but, whether mission impossible or not, you have to do a lot of ground work. For example, it seems you had not heard about ChessX GUI/Database and Lucas Chess. For that you have to be more broadly involved with chess. You have to also try out the various services and products to determine what they are lacking, before you make a pitch to VCs to secure funding. VCs considering funding would know a lot more about chess, game industry, and apps industry in general.
-
jefk
- Posts: 762
- Joined: Sun Jul 25, 2010 10:07 pm
- Location: the Netherlands
- Full name: Jef Kaan
Re: AI to explain chess positions to amateurs
well CK the chessgpt effort (as was given as answer to the OP) looks reasonable
https://github.com/waterhorse1/ChessGPT
while the LLM's will improve in reasoning (and logic) and possiby with some search, the project may
indeed become something. And it's not a commerical project (neither was the OP question i think).
The project on which Arun Mehta now seems to be working for more than a year or so (chessx which he probably
will rename) is another thing, while apparently he likes to spend some bucks on it, is still seems to be with
commercial intent and thus chances of success (either with VC's or not) imo are (very) slim, even if he would
start using this chessgpt. But that's all up to him, ofcourse (*).
But even then there are lot's of mobile (android) chess apps which *do* work (although i don't know
because i don't like the small screens so i'm glued to my desktop (and mostly win10).
(*) if it's mainly chess philantropy, spending 5/k month, then i would spend it on eg. Banksia,
making it more stable (it seems to be crashing quite often, but some of the features
as WDL (don't know if it has a vertical bar for that) and others apparently are quite kewl
https://github.com/nguyenpham/Banksia
- also a lot of work can still be done in the areas of opening names, and more sophisticated opening books
(than polyglot) and opening book (variation) training (like CPT which isn't maintained anymore)
https://sourceforge.net/projects/bookbuilder/
- in the area of chess training (kids or not) after the excellent and useful (**)chess tutors 1/2/3
(for reasons unkown to me) there isn't a chess tutor 4 (or possibly 3b) maybe the
market for that at higher levels is too small (anyway it's commercial):
https://www.shredderchess.com/windows/chess-tutor.html
(**) based on the well thought out chess-educational Dutch steps method
https://www.stappenmethode.nl/en/index.php
https://www.stappenmethode.nl/en/step4.php
https://github.com/waterhorse1/ChessGPT
while the LLM's will improve in reasoning (and logic) and possiby with some search, the project may
indeed become something. And it's not a commerical project (neither was the OP question i think).
The project on which Arun Mehta now seems to be working for more than a year or so (chessx which he probably
will rename) is another thing, while apparently he likes to spend some bucks on it, is still seems to be with
commercial intent and thus chances of success (either with VC's or not) imo are (very) slim, even if he would
start using this chessgpt. But that's all up to him, ofcourse (*).
agree, maybe the specific apps i mentioned (for kids) are not working as mobile app, didn't check it.You have to also try out the various services and products to determine what they are lacking
But even then there are lot's of mobile (android) chess apps which *do* work (although i don't know
because i don't like the small screens so i'm glued to my desktop (and mostly win10).
(*) if it's mainly chess philantropy, spending 5/k month, then i would spend it on eg. Banksia,
making it more stable (it seems to be crashing quite often, but some of the features
as WDL (don't know if it has a vertical bar for that) and others apparently are quite kewl
https://github.com/nguyenpham/Banksia
- also a lot of work can still be done in the areas of opening names, and more sophisticated opening books
(than polyglot) and opening book (variation) training (like CPT which isn't maintained anymore)
https://sourceforge.net/projects/bookbuilder/
- in the area of chess training (kids or not) after the excellent and useful (**)chess tutors 1/2/3
(for reasons unkown to me) there isn't a chess tutor 4 (or possibly 3b) maybe the
market for that at higher levels is too small (anyway it's commercial):
https://www.shredderchess.com/windows/chess-tutor.html
(**) based on the well thought out chess-educational Dutch steps method
https://www.stappenmethode.nl/en/index.php
https://www.stappenmethode.nl/en/step4.php