The Context Window
Your AI isn't forgetting you. It literally can't see past a certain point.
Grounded in the research on large language model
What it actually is
Every language model has a context window — a hard cap on how much text it can process at once. GPT-4 launched in 2023 with 8,192 tokens. Claude 3 ships up to 200,000. One token is roughly four characters of English, so 100k tokens is maybe 75,000 words — a short novel. Anything outside that window doesn't exist to the model. It can't retrieve it, can't reference it, can't be influenced by it. It's simply gone.
Where this came from
The constraint is architectural, not a design choice anyone can easily opt out of. The transformer architecture — introduced by Vaswani et al. in the 2017 Google paper 'Attention Is All You Need' — computes attention across every token pair in the input. That's O(n²) compute. Double the context, quadruple the cost. For years this kept windows small. Anthropic's constitutional AI work and sparse-attention research pushed the ceiling higher, but the wall never disappears.