The "next-token predictor" is taking a design stance (see Dennett). But to take the intentional stance is more interesting in case of large language/reasoning models.

[dead]