Language models play chess, everything is recorded (chessmark.merope.dev)

A new advancement in AI chess gameplay introduces a system where language models engage in chess matches while recording every action and reasoning process in real-time. The framework allows models to maintain a complete transcript throughout the game, tracking every move and rationale, which can be replayed to analyze decisions. This feature is critical for enhancing the transparency and interpretability of AI strategies in chess, as it enables developers and users to scrutinize the reasoning behind each move over an extended series of turns—up to eighty. Significantly, various AI models are competing, including Perplexity's Decider V1.1 and OpenAI's GPT-6 Luna Decisions, contributing to a growing landscape of AI chess engines capable of learning and adapting strategies dynamically. By leveraging comprehensive game histories and detailed positions represented via FEN notation, these models aim to optimize their gameplay through thoughtful analysis. The competition not only promotes innovation in AI-assisted chess but also stimulates discussions around the limits of machine learning in strategic thinking and decision-making in complex games.