Day 10/365 — Why AI forgets what you said ten minutes ago
You know the moment: at the start of a chat you give the AI a clear rule — "always reply in German" or "call me Stephan, not user." Fifty messages later, it calls you "user" again. In English.
This isn't the AI being stubborn. It's the context window.
What is the context window?
The context window is the model's short-term memory. It's the amount of text (measured in tokens) that the model can "see" at once. Everything outside that window is invisible. When the window fills up — usually after a few dozen messages — the oldest content gets pushed out.
Why this matters
The context window determines what the AI knows about you, your conversation, and your instructions. If you've given important context early in the chat, it will eventually be forgotten. This isn't a bug — it's the architecture.
What you can do
- Put important instructions at the END of long conversations (or near the end)
- Use system prompts for permanent rules
- Summarize long conversations periodically
- Know your model's context limit (Claude: 200K, GPT-4: 128K, Gemini: 1M)