Context windows limit short-term AI memory

A model can only process a finite number of tokens simultaneously, causing it to ‘forget’ earlier parts of extensive prompts. The principle that Working memory is severely limited applies equally to silicon and biological brains.¹

Footnotes

1. OpenAI. (2023). GPT-4 Technical Report. OpenAI. ↩