Memory layer

The memory layer is what a model carries from training: the impression of you already inside it before any search happens.

Used in this book as one half of the pair that explains where an AI resume comes from.

Elsewhere it is usually described as parametric knowledge or model weights, which names the mechanism rather than what it does to a reputation.

The memory layer is fixed until the next model release. The retrieval layer changes as soon as the sources do.

A correction to the memory layer cannot be filed. It has to be earned, through enough consistent public evidence that the next training run reads it differently.

The book's practical point is that the memory layer needs patience and repetition rather than a request, and that it updates on release cycles you do not control at all. It is the reason the book insists reputation moves on machine time rather than campaign time.

Read about the book's argument.