skip to content
Large Language Muddle

The context window, or: a goldfish with a library card

/ 1 min read

View more blogs with the tag sample, View more blogs with the tag ai, View more blogs with the tag explainer

PLACEHOLDER: dummy copy to show the tone. The explanation is deliberately simplified. Replace or delete.

Every chat with a language model happens inside a context window. Think of it as a desk. Everything you’ve said and everything it’s said has to fit on the desk. When the desk is full, something falls off the edge. Usually the bit you needed.

So the model isn’t ignoring you. It’s a goldfish with a library card: brilliant access to an enormous collection of stuff, and no idea what you asked three minutes ago if the conversation has gone on a bit.

A few things I’d like to understand properly, rather than nod at:

  1. What actually counts towards the window (it’s tokens, not words, which is its own little scandal).
  2. Why long conversations get sloppier.
  3. Whether “just start a new chat” is wisdom or giving up.

I’ll report back. The desk is clear. For now.

AI up me duck