AGI Maze as a Benchmark Framework for World-Modeling Agents

Alexey Potapov

Large language models (LLMs) are powerful pattern-completion systems, but their default operating mode - predicting the next token from a static context - does not reliably produce…