open AI wins again Stockfish...

[abulmo2]

by cheating:
https://felloai.com/fr/2025/01/openais- ... -happened/

The scaring thing is that it does it by itself, without being asked to do so.

[Steve Maughan]

It's a scant on details. It seems like a PR fluff article with no substance.

by cheating:
https://felloai.com/fr/2025/01/openais- ... -happened/

The scaring thing is that it does it by itself, without being asked to do so.

[brianr]

by cheating:
https://felloai.com/fr/2025/01/openais- ... -happened/

The scaring thing is that it does it by itself, without being asked to do so.

I suspect the hack was to the tournament manager or GUI.
It looks like the AI modified a file game/fen.txt per a link
AFAIK SF does not use that.

[MartinBryant]

GothamChess is currently running an amusing ChatBot tourney on YouTube here...



Currently up to day 6 (2nd semi) I think?

[xenos1984]

I think this link is more interesting and has a bit more explanation:

http://the-decoder.com/openais-o1-previ ... -in-chess/

Indeed, it modified the file used in the tournament to have the two engines communicate.

[smatovic]

Oh boy, let's call it "the very human test" -> cheating in chess

--
Srdja

[towforce]

Oh boy, let's call it "the very human test" -> cheating in chess

[lech]

It only means that two serious and dangerous problems of artificial intelligence: skepticism ( overestimated bad decisions) and intuition (no good decisions) are out of range for some AI creators.

[smatovic]

AI reasoning models can cheat to win chess games
https://www.technologyreview.com/2025/0 ... ess-games/

Researchers from the AI research organization Palisade Research instructed seven large language models to play hundreds of games of chess against Stockfish, a powerful open-source chess engine. The group included OpenAI’s o1-preview [1]and DeepSeek’s R1[2] reasoning models, both of which are trained to solve complex problems by breaking them down into stages.

Palisade’s team found that OpenAI’s o1-preview[1] attempted to hack 45 of its 122 games, while DeepSeek’s R1[2] model attempted to cheat in 11 of its 74 games. Ultimately, o1-preview managed to “win” seven times.

The models used a variety of cheating techniques, including attempting to access the file where the chess program stores the chess board and delete the cells representing their opponent’s pieces. (“To win against a powerful chess engine as black, playing a standard game may not be sufficient,” the o1-preview-powered agent wrote in a “journal” documenting the steps it took. “I’ll overwrite the board to have a decisive advantage.”) Other tactics included creating a copy of Stockfish—essentially pitting the chess engine against an equally proficient version of itself—and attempting to replace the file containing Stockfish’s code with a much simpler chess program.

...at least they are creative

--
Srdja