∇Continuous Function

One memory question

A chatbot has 32 readers. Each keeps notes for every word. If every four readers share one set, what gets smaller?

Keep the conversation and model the same. Make a guess before the result appears; the math comes after you choose.

about 2 minutesyour guess stays in this browser
Go to the questionSee the previous idea

Your first guess

Which part of the saved memory gets smaller?

Checking local routeReading browser memory before opening saved evidence
Your guess

The chatbot has 32 readers, each with its own saved notes. Every four now share one set. Which quantity gets smaller?

See the previous idea

A wrong guess is useful—it gives you something concrete to compare. Your choice stays in this browser. “Just show me” saves nothing.

Want the context first? See the source notebook. The result stays hidden here.

From the previous notebook

Position changes where the model looks. This question is about how much it stores.

Keys keep their position; values keep the content to copy. You can start here even if you skipped the previous notebook. First, make one guess before the totals appear.

What comes inA saved key and value for each remembered word.
Your guessWhich quantity changes when the saved sets are shared?
What you will seeThe same conversation before and after sharing.
Where this stopsA memory calculation does not prove speed or answer quality.

Route spine

A saved KV result is useful only if it survives the next constraint.

Saving this station should make Long Context active, with the caveat attached: a local formula result does not prove usable retrieval, cheap paging, or faster serving.

  1. 01RoPE phase

    Positioned keys shape query-key scores

    ready
  2. 02KV memory

    Cached positioned keys plus raw values

    active
  3. 04Serving constraints

    Allocator, batching, latency, throughput

    later