The Imprinting Window (AI)

In Konrad Lorenz’s ethology, an imprinting window is a narrow developmental period of maximal plasticity during which a specific and consequential learning occurs—after which that learning is irreversible in ways that all subsequent experience cannot undo. A greylag gosling imprinted on a human cannot be re-imprinted on a goose by wishing or by prolonged exposure; the first impression is not a draft to be revised but the foundation on which later experience builds. The AI imprinting window maps this concept onto large language model pretraining: the window of radical parameter plasticity that opens once at the start of training and closes as the corpus is absorbed, setting a structure of representations, associations, and biases that fine-tuning and alignment can shape but not rebuild. The corpus a model meets in this window becomes, past appeal and past revision, the standing pre-reflective answer to what the world is like—what words mean, what follows what, what coheres and what cannot. The