AI to explain chess positions to amateurs

[Player_one]

When will there be a decent LLM or GPT that can explain chess positions? We're not that far away.
Is anyone working on this? I would like to help, if I can.

I saw 6. a3 as the recommended book move in the line below and had no idea why White would "waste" a tempo.

So I asked chatGPT:

In the following line from the French Defense 1. e4 e6 2. d4 d5 3. e5 c5 4. c3 Nc6 5. Nf3 Qb6 , why would White want to play a3 next? What strategic ideas from the Black side is the move attempting to prevent?

Gemini's answer was problematic as it confused light squares and dark squares but I could see where it was going.

ChatGPT's answer left something to be desired but it was helpful to me.

In the given position from the French Defense:
e4 e6
d4 d5
e5 c5
c3 Nc6
Nf3 Qb6
White might want to play 6. a3 next to prevent several strategic ideas from Black:

1. b4 Square Control: By playing a3, White gains control over the b4 square, making it difficult for Black to expand on the queenside with moves like ...b5 followed by ...b4. This also prevents Black from harassing White's pieces, particularly the c3 pawn and the knight potentially landing on b4.

2. b4 Push: It stops Black from playing ...b4 themselves, which could be used to undermine White's center or to gain space on the queenside. The move a3 ensures that White's pawn structure remains intact.

3. Preparing b4: In some variations, White might aim to play b4 themselves, gaining space and creating a potential pawn break. The move a3 supports this idea.

4. Preventing ...Na5: While not immediately a threat, the move a3 also takes control of the b4 square, which could be useful in preventing Black's knight from jumping to a5 and attacking the c4 square after White eventually plays dxc5.

Playing a3 is a prophylactic move that strengthens White's position and prepares for future pawn advances while limiting Black's counterplay on the queenside.

[isidro02139]

Hey @Player_one – I'm not building an LLM, but I am working on a modern mobile game that teaches chess with AI-powered characters (e.g. the Seamstress teaches pins, the Gambler teaches sacks, etc.). The AI is hand-coded algorithms combined with open source engines, and I am trying to recruit some cool chess programmers to help me build the app which I call ChessX for now.

Would you have any interest in helping me with deep, specific AI algorithms that we are building to support these unique AI teachers?

Thanks! Also, maybe you know a better person (or forum) to ask, sorry in advance!
Arun (arun@chessx.app)

[chesskobra]

The programmers here are doubleplus uncool, so they are not likely to work on such fun apps. From the description, it sounds like a VC funded company that wants to teach kids to play like Tal in a fun way.

In case you have not heard, chessx is a well known chess GUI. So you may want to call your app something else.

Also, if you post the same message in multiple unrelated threads, you will be reported.

[isidro02139]

> The programmers here are doubleplus uncool
Awww... well maybe someone is interested in making games for a new chess audience

> sounds like a VC funded company
Unfortunately, no funding (yet!) I'm just bootstrapping it for a few years

> chessx is a well known chess GUI
Ah, didn't know! I like the 'X' as I search for the ideal name (should connect to the final meta)

> if you post the same message in multiple unrelated threads, you will be reported
Thx for the tip!

[chesskobra]

I may sound cynical. But how much market is there for the app that you are proposing to develop? AI character explaining pins? It sounds like creating an app or a game that teaches multiplication tables of 2 to 10. You would say there is more to it than just pins, but I think ultimately you would be only scratching the surface. I don't use such apps, but I suspect most such apps are hardly able to go beyond scratching the surface. Also, if one plays a bunch of games, one quickly goes beyond understanding what a pin or fork is, or mate with king and queen. Kids' parents are better off hiring a human coach, who does not have to be a titled player.

There are also excellent programs (apps) and sites for training such as Lucas Chess and Lichess, which are free, open source, feature rich, and very actively developed, although they don't market their features as 'AI Coach'. Companies like chess.com, chessbase, etc. may already be doing a lot on all aspects of chess training.

Another practical issue you would have is what kind of programmers can you afford to hire? Some programmers on this site may be very elite programmers (I am not one of them). They know chess and programming, and many of them would also have high level formal education. You are not going to get that kind of programmers.

On the other hand, if you can really address what OP is asking, for example, AI that can explain positional nuances, that will have much bigger market of intermediate players, and will also attract better programmers. For developing such a product, a startup founder himself or herself would have to know both chess and programming at an advanced level.

[j.t.]

When will there be a decent LLM or GPT that can explain chess positions? We're not that far away.

Take a look at https://github.com/waterhorse1/ChessGPT

[isidro02139]

> I may sound cynical. But how much market is there for the app that you are proposing to develop?
Not at all! I think of this game as a casual version of chess, we are targeting existing casual gamers (like, match-3 gamers – the lowest of the low... lol!)

> ultimately you would be only scratching the surface
It's challenging for sure – level 1 of each AI piece shows the most basic version of the concept (Bishop pin), and we try to expand it up to level 4 (pins for all piece types, specific tactics around pins etc.) Each piece tries to focus on a single concept, and the combination of pieces allows the player to see different trade-offs and opportunities in our AI categories of Time, Strategy, Tactics, Defense, and Openings (also have a Meta category for things we can't box cleanly yet!)

> Kids' parents are better off hiring a human coach, who does not have to be a titled player.
One goal is to make it free to learn for everyone, also to let people have a 'casual' chess experience (those that would not want to explicitly think of themselves as 'chess players' or interested in active learning).

> There are also excellent programs (apps) and sites for training
True, but I haven't seen anything exciting in a while – part of why I'm' making this app is I want it for myself!! Thx for the Lucas Chess reference

> what kind of programmers can you afford to hire?
I'm spending about $5K a month, that covers EVERYTHING – art, programming, etc. (so, not a lot.. ). But, I have found some great programmers in the past that were willing to work part time or even for free because they are passionate about either AI or chess/chess programming (legit smart people).

> AI that can explain positional nuances will have much bigger market of intermediate players
Developing an LLM would be cool, but out of the scope of this project and my technical background (though I am a programmer). I hope our game will be able to deliver such nuances through our Stockfish-aided custom algorithms and arrows on the board.

Thanks, chesskobra, for all your helpful thoughts & advice!

[towforce]

Just "thinking out loud" about this: what do you tell your customer about a position?

Same applies to human trainers: there's a lot going on in most positions. Maybe the REAL skill of a strong player is knowing what to ignore, so that their thinking can be directed only to what's important.

It might be good for a weak player to be given a list of aspects about a given position, ordered by importance - with an explanation as to why each aspect gets the importance rating it's given in this position.

You're spending $5k a month on this - where's that money coming from?

[shawn]

There is an open source project called ChessCoach that aims to do this, but I have no idea how it works and how well it is at explaining chess moves.
https://github.com/chrisbutner/ChessCoach

[jefk]

besides this chess coach there is already http://decodechess.com
(more than a year ago it (the explanations) was/were too simple for me, but now you can play a game
online against the comp; so i choose level 2 and at move 8 the app resigned against me; when i asked
explanation they ask me to pay). The Fritz19 sometimes does some decent explanation (with or without
'buddy' engine although i sometimes have some issues with it (eg. quickly analyzing a game).
chess software for kids, well there is http://chesskids.com
and Fritz and Chester by Chessbase https://shop.chessbase.com/en/products/ ... d_chesster
And I recently i came across another app apparently also chess training for kids
https://www.chessmatec.com/
So in other words, there are plenty of such apps. If you have some AI algorithms for use of the above
apps then i would approach them and see if they are interested (to pay or not); probably not unless
you really have excellent communication skills (in this subculture) like indeed (ck) often is not the strongest
point of many programmers in this -technical- field. And if you want to compete with them from scratch well good
luck (mission impossible imo in commercial chess software there's chessbase and chess.com lots of open sources
software so other commercial software can't compete; mr ckobra seems to express himself in a cynical
(or sarcastic) way sometimes (with an intention simply to be honest i suspect) but this time imo he certainly
was 100 pct right; there's an overload of chess information nowadays eg on youtube or elsewhere
and a commercial attempt in this area simply won't be a smart endeavour.

PS as for the chatgpt explaining that 6.a3 move (which is good indeed in that advance French variation), it said 'preventing
b4 by Black) HAHA, there is a queen on b6 and the Black pawn is still on b7. As for preventing Na5, also bs, in this var Na5 is
played quite often (sometimes first c4) and then Na5 b3), so it's not preventing Na5, it's inviting Na5.
In otherwords the usual Chatbs and of no use in chess explanation(s) at all. Ask an IM and he can explain it to you within
a minute (i can explain it as well but then I lookup some GM games and this would take a few minutes). The move
6.Bd3 usually is sharper eg leading to the Milner Barry gambit in old style or more advance form, after 6.Be2 also a
good move, d4 becomes weak and Black can play Bb4, so 6.a3 is good positional move, for several reasons combined
(let's say it's just opening theory). PS2 ok being curious just took 2 minutes to research the games, while it's often played
by GM's, in 2022 it was in the opening list of the TCEC Premier league, and the games KomodoDragon2-StoofvleesII
and StoofvleesII- Komododragon2 both ended with a win for White in this variation. Some superGM's
playing the move include names as Firouzja, Nepo, MVL, Giri, Caruana, Grischuk etc. So it
can't be a bad move that's for sure