If you mean within an individual context window, yes, that's a known phenomenon.

If you mean over the lifetime of a model being deployed, no, that's not how these models are trained.