Chess System Tal 3 released

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

User avatar
chrisw
Posts: 5137
Joined: Tue Apr 03, 2012 4:28 pm
Location: Digital Nomad. Anywhere but the Western Empire
Full name: Christopher Whittington

Re: Chess System Tal 3 released

Post by chrisw »

Ipmanchess wrote: ↑Sat Oct 03, 2026 5:53 pm Well done could enter main-list:

Chess-System-Tal 3.00 avx2 added to R9 list: https://ipmanchess.yolasite.com/r9-7945hx.php

OK: No time forfeits found
OK: No disconnections found
OK: No illegal moves found

Ipman.
Thanks for testing
Chessgate
Posts: 33
Joined: Mon Mar 09, 2026 12:43 am
Location: Rangsdorf , Germany
Full name: Arne Bailliere

Re: Chess System Tal 3 released

Post by Chessgate »

Morphy Elap wrote: ↑Sat Oct 03, 2026 4:50 pm
mclane wrote: ↑Sat Oct 03, 2026 2:25 pm
Morphy Elap wrote: ↑Fri Oct 02, 2026 3:35 pm May I know how it performs in EAS? Besides, is this version the most closest to Mikhail Tal? By any chance, can make other variations for morphy or Anderssen or other romantic era masters? Thank you very much.


This question is solved:

https://playchessgate.com/
Sorry to have opposite stance with you, indeed it tried to immetitate the player with opening books and style like attack, positional through selection of multiPV, but it failed to understand the core principles of that player and doesn’t know how much chance that player prefers on that move and what does that person think about the position. Also playchessgate engine just never make simple cleaners, but even top level player might occasionally make the slightest error by overlooking smth, a random blunder made by engine will not make sense. Instead I actually trained networks of the romantic era masters from their games based on Nova chess ai (trained on human games with styles) (a policy head), to predict the player’s move according to the chess knowledge and style of the players base on its games (so that have same style even the player has not met that position). The only problem is that the strength is only roughly 2700 blitz, but in classical match, it performs like 2000 cause it doesn’t even calculate and just pretends to be a human. Therefore, I trained a value head with PUCT search just like Lc0 for Morphy personality and now other in progress. In my own developing app, I can now just learn from those players’ moves. The only problem is it doesn’t have the tactical ability of chess system tal especially extreme. So my lines prepared by my bots will just be a positional concept but with lack of calculated sacrifices. I hope to have a personalized personality mixed with chess system tal to create a unique personality with huge search and knowledge base, but not only sorting those top moves.
Thank you for using Chessgate and for your feedback . If you are interested in the deeper insights of how the system works there is the page "how the AI works" . However, this thread is about Chris ' s program which i highly respect. So i will not enter a discussion here 🙂 Best regards