Real game AI, not a chatbot: why these opponents don't use an LLM
Syndicated from the original on lkforge.com . The engines are playable in your browser at lkforge.com/games ; the harness that produced these numbers is public and seeded. Every "AI" in a product now seems to mean a large language model. The AI that plays against you on my site doesn't — it's classical game-tree search: minimax, expectimax, breadth-first search. That's a deliberate engineering…
Modern AI opponents on gaming websites do not utilize large language models (LLMs) to make their moves. Instead, they rely on classical game-tree search techniques such as minimax, expectimax, and breadth-first search. The AI employed on the reporter's site demonstrates a deliberate engineering choice, aiming for provable correctness and sub-frame latency without involving a server.
This approach differs from using an LLM, which is trained to predict the next token of text rather than searching a game tree. While LLMs excel at generating fluent text, they cannot provide the determinism, correctness proof, and rapid latency that game-tree search algorithms offer. The reporter's tic-tac-toe engine, for instance, returns a provably-optimal move in approximately 0.3 milliseconds on the user's device, with zero network calls and 0 losses in 1,200 test games.
This deterministic performance contrasts with an LLM, which might generate a winning prompt but could still make suboptimal moves due to its training objectives. The reporter argues that for board games with fixed rules and finite trees, classical search is the ideal tool rather than the trendy LLM.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
