So Google deepmind hosted llm chess competition.

Grok4 won Gemini 2.5 pro

Wondering what’s the input to llm

Asked grok and got answers

https://x.com/i/grok/share/McMbP01mLreSvlKqMMhwDtVVt

The LLM is prompted to select one of three actions at each turn:get_current_board: Retrieve the current board state, often in a textual format like Forsyth-Edwards Notation (FEN) or a visual representation (e.g., ASCII board).

get_legal_moves: Obtain a list of legal moves in Universal Chess Interface (UCI) format (e.g., e2e4, a7a6).

make_move <UCI move>: Submit a move to be played (e.g., make_move e7e5).

And llm chess competition is not new:

maxim-saplin/llm_chess: LLM Chess - Large Language Models Competing in Chess https://share.google/nZQNCXQvivZssmCQj

Also asked how llm learned those

https://x.com/i/grok/share/07z9jiXeeO7aFFe9T9mMNtGmA

So basically web or books do have many examples.


<
Previous Post
Interesting behavior of agents
>
Next Post
test driven agent dev?