Skip to content
Notes on AIE014Act 2 — Behavior & Limits

Context Window

E014 · by — AI Strategist, Practitioner & Educator · Founder, AI Center of Excellence

Define what the model can and cannot see at once.

Full Explanation

The context window is the amount of information a model can see at one time. It's not memory — it's working space. The model can only reason about what is currently visible. At any moment, it receives a fixed block: your message, some recent messages, some instructions. That block is its entire world.

If something is inside that block, the model can reason about it. If something is outside, it does not exist — not ignored, not forgotten, just invisible.

A useful metaphor is a desk. Only the papers currently on the desk can be used. As new papers arrive, old ones slide off the edge. Once they fall off, the model cannot look at them anymore.

This explains why instructions seem to disappear, why answers contradict earlier statements, and why long conversations slowly fall apart. The model isn't being careless — it simply no longer sees what you think it should remember.

The key insight: reasoning only happens over what is visible. The model does not reason over the past — it reasons over the current context.

Transcript

Read every word ↓

AI doesn't forget what you said, it just stops seeing it. Most people think AI remembers the whole conversation. Normally it doesn't. What actually matters is something called the context window. The context window is simply the amount of information the model can see at once. Not the full long chat, not everything you ever wrote, not all the attachments, just a limited slice of text data that is currently visible to the model. And here's the key idea. The model can only think about what it can see. At any moment the model receives a fixed block of context.

Your current message, your recent messages, some instructions, some other information. This block for the model is its entire world. If something is inside that block, the model can reason about it. If something is outside that block, it doesn't exist. Not ignored, not forgotten, just invisible. A useful way to think about it is a desk. Imagine a model has a desk in front of it. Only the papers currently on the desk can be used. As the new papers arrive, all the ones slide from the edges and disappear.

Once they fall off, the model cannot look at them anymore. It doesn't matter how important they were or how correct they were or how carefully explained to the model information on this piece of paper. If they are no longer visible now, they cannot affect the answer. That's what the context window really is. It's not intelligence, it's not memory, it's the working space, it's this desk. And this explains a lot of strange behavior of AI models. Why instructions seem to disappear, why answers contradict the early statements, and why long conversations slowly fall apart.

The model isn't being careless, it just simply cannot see what is no longer in the context window. The key insight is the following one. Reasoning only happens over what is visible. The model doesn't reason over the past. It reasons only over the current context window. Once you understand that, many AI features stop feeling mysterious. They start feeling mechanical and manageable. This was a note on AI for today. Thank you for watching and I'll see you in the next one. Bye.

Resources

No dedicated resources for this episode yet.

Browse the resource library →

More from Act 2 — Behavior & Limits

Newsletter

Found this useful?

Join the newsletter. You get one e-mail when something new and useful is out: a video, a guide, a tool. No spam, no schedule, and you can leave with one click.

Alexey Makarov

Alexey Makarov

AI Strategist, Practitioner & Educator. Founder & Global Lead of the AI Center of Excellence at SEFE. Creator of the Unreasonable AI YouTube channel. Based in Berlin.

About Alexey →