LANDMARK · 5 MIN · ORIENT
The context window
After this landmark, you can predict when a model will “forget” or miss something, because you know it only works with what’s in its window.
A model’s context window is its working memory for one exchange: your prompt, the conversation so far, and anything you pasted, all measured in tokens. It has a hard maximum. Anything outside it — a document you didn’t share, a detail from far earlier in a long chat — simply doesn’t exist for the model. When answers quietly drop earlier detail, this limit is usually why.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
In a long conversation, the model can lose the thread from the top. If it matters, restate the key detail rather than assuming it still “remembers”.
MAKE A DECISION
Halfway through a long chat, the model contradicts something it agreed to near the start. Most likely cause?
CARRY THISIn your next long AI chat, restate the one detail that must not be lost right before the question that depends on it.