Is there any way of combining ChatGPT with AlphaZero?

@CheerUpChess-Youtube said in #8:

Yes that would already be great! This way I could analyse a lot of games far more easily. I have a very big project in mind. So you are saying it doesn't exist? Is there any way for me to add chess engine evaluations and commentary to a png like (white is better because of king safety) - claim made by a chess engine?

I am not saying it doesn't exist, but probably doesn't exist in a polished form (with text annotations like GM annotations). You can already add engine evaluations in a pgn automatically (look at Arena or Scid vs PC). These evaluations are just single numbers (in most tools). I have seen a little more than that using a python chess library, where for a given position you could print a summary table of various features like material, space advantage, king safety, mobility, advanced pawns, ... as separate evaluations. In fact I think such a thing may be easier (and preferable) to do with some older engines (e.g., crafty, phalanx), which use classical evaluation. and have simpler C code, instead of stockfish which is complex. The only disadvantage of using these engines will be that they may not have UCI interface, so the python library may not be able to interface with them, and one will have to directly deal with the engine code. But the problem is not much more difficult than understanding the relevant code and adding a bunch of print statements.

Incidentally, there is a paper by Berliner, Kopec, and Northam (1990) Taxonomy of concepts for evaluating chess strength, which gives a very detailed list of features of a position. One needs to print out evaluations of these features from engine analysis.

@petri999 said in #7:

Traditional chess engine is AI. not all AI is machine learning let alone deep neural networks. One first trials on AI was chess playing programs.

Of course, you are right. I meant in the sense - you don't need "hyped up AI" for that. In the 70s and 80s, many people, including Danny Kopek and Hans Berliner (both strong chess players and computer science professors - Kopek was IM, Berliner was correspondence chess champion) were heavily involved in this. It is worth looking at their paper, which I mentioned in another comment.

and chat GTP would produce some semisensical story out of any position as it has enough chess related chat fed in during training phase. whether that would offer any benefit at all is another issue

Again, I agree. I would say "semisensical and feel good story" like, for example, "Ba7! Karpov is obviously thinking like a boa constrictor here". I doubt chatGPT like stochastic parrot will be very useful. But it might sell better if for example chessbase adds support of it. For all you know, they may even be trying.

@kajalmaya said in #11:

But the problem is not much more difficult than understanding the relevant code and adding a bunch of print statements.

.... man I want to just have someone say "this is a good move" "a very good move" or "a bad move" depending on the change of evaluation provided by stockfish. There needs to be an easier way to do this. It's so simple. If (lower than before) say "bad move" and so on. I could code that with ifs, but firstly I would need to know how to export a pgn from lichess that only has evaluations in it and not variations and also how a program could read exactly that data out. I have no clue how parsers work. Please help. Your answer still is very much appreciated. I wanted to learn python, but I feel it's overwhelming and I don't have the patience. I need an easy solution for this.

@CheerUpChess-Youtube said in #13:

.... man I want to just have someone say "this is a good move" "a very good move" or "a bad move" depending on the change of evaluation provided by stockfish. There needs to be an easier way to do this. It's so simple. If

There is python code to generate puzzles out of pgn games - for example, you set a threshold on the evaluation and if there is a change in evaluation that is more than the threshold, then it outputs the position. Of course, you will have to learn a little programming and be comfortable running such tools on the command line.

@kajalmaya said in #14:

There is python code to generate puzzles out of pgn games - for example, you set a threshold on the evaluation and if there is a change in evaluation that is more than the threshold, then it outputs the position. Of course, you will have to learn a little programming and be comfortable running such tools on the command line.

Hmm okay can I see this code to see if I understand it? So basically I need to write a code to do what I want to do?

https://github.com/vitogit/pgn-tactics-generator
just one on many. I found 13 of this type in github alone. Problem is that how you know which one is good.

All good problems sets - like in chess tempo - are curated. Fully automatic tends have so really silly problems. or ones where solution is not understandable. i.e you get winnin position with equal material and you need find all by yourselves why it is winning.

It will take some effort to test if any of those is to you liking. But the linked has more "stars" than the others

@petri999 said in #16:

github.com/vitogit/pgn-tactics-generator
just one on many. I found 13 of this type in github alone. Problem is that how you know which one is good.

All good problems sets - like in chess tempo - are curated. Fully automatic tends have so really silly problems. or ones where solution is not understandable. i.e you get winnin position with equal material and you need find all by yourselves why it is winning.

It will take some effort to test if any of those is to you liking. But the linked has more "stars" than the others

Okay I don't understand anything written in this code lol. I really need another approach to this than a python script. Could I somehow use excel for this task? Can excel read out and create files or not? Because it certainly can calculate something and link a response to a certain result.

Excel is in compurter science parlance "turing complete" yes it can BUT once you start to do anything truly complex it becomes impossible to understand or maintain.

But it can read files and it has functions you can define so you can do almost anything with it. Just it rarely very good solution

@petri999 said in #18:

Excel is in compurter science parlance "turing complete" yes it can BUT once you start to do anything truly complex it becomes impossible to understand or maintain.

But it can read files and it has functions you can define so you can do almost anything with it. Just it rarely very good solution

How can I import a PGN, change it and export it again this way?

It is not that different from python so it is not going much easier
https://www.automateexcel.com/vba/read-import-text-file/
basically you write VisualBasicforApplication function that read the file and then parses ans puts on the spreadsheet
Sub FSOPasteTextFileContent()
Dim FSO As New FileSystemObject
Set FSO = CreateObject("Scripting.FileSystemObject")
Set FileToRead = FSO.OpenTextFile("C:TestTestFile.txt", ForReading) 'add here the path of your text file

TextString = FileToRead.ReadAll

FileToRead.Close

ThisWorkbook.Sheets(1).Range("A1").Value = TextString 'you can specify the worksheet and cell where to paste the text file’s content

End Sub

and writing would be something similar.
Once start doing bit more complex stuff with excel it gets closer to normal programming. As said can be done just rarely worth the trouble.

Sometimes if you data is in simple text format in field with comma/semicolon/colon separation you can use the excel import csv file to get you data. In you case you need produce that text file and I really dont see any other way than writing it in some real programmin language like python.