What counts as an AI world model? A post argues prediction and planning aren’t the same

A post argues that generating a plausible future, predicting what an action will cause, and helping choose a better action are distinct achievements. Calling them all “world models,” the author says, shouldn’t blur what each has demonstrated. They also argue that agents should compare possible futures before committing, and that models should connect imagined outcomes to concrete actions while retaining the details that matter. For opening a drawer, they suggest, the latch matters more than the wood grain.