Context window

The maximum number of tokens the model can see at once. GPT-1: 512 tokens. GPT-2: 1024. GPT-3: 2048. Modern models: up to 1 million+.

The maximum number of tokens the model can attend to at once. GPT-3's context window is ~2,000 tokens (roughly 1,500 words). Longer documents must be truncated or split into chunks.

The maximum amount of text (in tokens) a model can consider at once — its short-term "working memory" or "desk."

The maximum amount of text (measured in tokens) a model can consider at once — its working memory. Fill it with the right things; when it's full, older content must be summarized or dropped.

The maximum amount of text (in tokens) a model can read at once — instructions, examples, and question together must fit inside it.

The maximum number of tokens a model can read at once (its short-term memory / desk space). Everything in the prompt counts against it. (Mod 1)

Copilot's finite short-term working memory (in tokens) for one session — prompt, code, instructions, skills, memories. (Mod 13)