Chess
2304 rapid on Chess.com, and 54th of 624 in the Tiny Chess Bot Challenge. The thread that connects the engine, the training platform and the evening games.
- Ongoing
Serious play since 26 February 2020. The program came before it: a first attempt at a chess engine in Python, in 2019. The program first, the game second.
Chess is a precise choice, not an accident: probably the best-known strategy game in the world, universal, without an ounce of luck, and with objective ways to measure the level of a player or an engine. That makes it an excellent test bench for what an AI is worth: complex enough to be a real challenge, discrete and finite enough as a space to remain tractable to program.
Most of it is played online, with no federation rating, no club and no official tournament. What can be checked in one click: 2304 rapid on Chess.com, also the best. Also present on Lichess.
Over-the-board play exists, but with no rated stakes: the EchequIIE association at ENSIIE, games at school, and more recently a few chess bars and games among friends. It is a different pleasure, where you can see the person think.
In 2023, Sebastian Lague’s Tiny Chess Bot Challenge required fitting an entire bot in 1,024
code tokens. Grogros (by: Grobert) finished 54th out of 624. The contest ranking measures what
you manage to fit within that constraint, not an engine’s strength; the main engine, outside the
contest, behaves differently.
Writing the engine taught chess as much as playing did. Writing an evaluation function makes you understand that a position is not worth plus one point two: it is won, drawn or lost, and the number only exists because we cannot see which. It changes the way you play.
It worked the other way too: an en passant capture suffered in a game, the rule reread properly, then added to the program.
What is most enjoyable is not theory or piece values: it is finishing a game cleanly, with continuity and an attack on the opponent’s king carried by sacrifices. Play somewhat in the romantic style, where the pieces are just tools to mate the king, and losing material in the battle is fine if it wins the war.

With rising level, fewer sacrifices and progress in strategy, tactics and calculation: a more solid, composed game. No single game or milestone to celebrate above the others. But the game is as mentally exhausting as it is satisfying, and time eventually ran short: the energy went to other projects. Today, hardly any games, and it is not missed.
It is the thread that runs through the rest of the site: the engine, the training platform and the game itself come from the same place.
