Skip to content

Knowledge Cutoff and Dual Memory

Entry #611 / 700 · Chapter 13 · 13.1 AI search products, RAG and query fan-out · Read in Chinese →
In one sentence

The two memory systems a model operates with: parameters fixed during training and frozen at a cutoff date, and a temporary working memory assembled from live retrieval at the moment of asking.

Analogy Mental model

A professor whose head holds everything learned before graduating, and who must put on glasses and read this morning's paper to answer anything about today.

In practice What you actually do

This splits the work in two. The tactical layer targets retrieval and moves in days: rank in the index that is queried, push updates, structure paragraphs properly. The strategic layer targets the parameters and compounds across generations: sustained presence in the large public corpora that the next model will be trained on.

Where people get burned Pitfalls

No amount of tag editing on your own site will change what a model already believes. Parameters respond only to accumulated presence in high-authority public sources over time.

This entry was written in English from the Chinese original, not machine-translated. Read the Chinese version →