AI models have a fixed limit on how much text they can hold in mind at once—the context window—which means long conversations fade and earlier nuances disappear, much like human short-term memory constraints. Understanding this limitation helps you structure prompts strategically: repeating key context, breaking large tasks into segments, and using explicit summaries to keep important details in focus.
Think of an AI's context window like a conversational workspace. Imagine you're working at a desk, but you only have space for a certain amount of papers in front of you at once. Everything you're actively working with fits. But if you try to keep 500 documents open on that tiny desk, you run out of space. AI context windows work the same way—they're the amount of conversation AI can 'see' and remember at once.
Here's what happens in practice: You're having a long conversation with Claude about your project. You spend 20 minutes discussing details. Then you ask a question about something you mentioned way back at the start. But 20 messages later, that information is outside Claude's 'visible workspace'—it's still in the conversation history, but Claude can't actively hold it in mind while processing your new question. So it might give you an answer that contradicts what you said at the beginning, not because it forgot, but because it literally can't see that far back while thinking about your current question.
Different AI tools have different context window sizes. Claude 3 has a large window (200,000 tokens—roughly 150,000 words). GPT-4 has a medium window (8,000 tokens for the base version, 128,000 for extended). Older tools might have much smaller windows. 'Tokens' are basically small chunks of text—a rough estimate is 750 tokens per 1,000 words, but it varies.
Why this matters for productivity: If you're working on a big project and having a long conversation about it, you eventually hit the limit. The AI can no longer reference early-conversation context. For practical purposes, you just see the AI's answers becoming less coherent or contradictory. The solution? Start a new conversation, briefly re-establish context, and continue. It's like closing your 500 open documents and keeping only the active ones.
A practical workaround: If you're doing long-term work with AI, periodically summarize what you've decided and paste the summary at the start of a new conversation. This keeps the AI 'oriented' without taking up window space.
Try this: In a long conversation with Claude (10+ back-and-forth exchanges), ask it to reference something you said five exchanges ago. It'll likely have it in context. Ask about something from 30+ exchanges ago. It might struggle. Then try starting a new conversation and pasting the early context at the top. Notice the difference in responsiveness.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.