The Desk, Not the Brain: Context Windows
By the end you'll be able to
- Explain the context window as the fixed token budget the model can use for the current response.
- Explain that "forgetting" mid-conversation is usually material dropping out of the window, not memory decay.
- Manage the window deliberately: summarise, restate, or start fresh.
- Predict
- Manage
- Decide
- Field note
⚡ 30-second warm-up — From Lesson 9: what gets assembled fresh for every single loop of generation?
The whole bundle of tokens — hidden instructions, the conversation so far, anything pasted — reassembled and re-read on every pass. Today you'll meet the hard limit on how big that bundle can be.
Twenty minutes into a chat, the AI "forgets" your name from message one — yet it used your name happily back at minute five. Find out why.
Start with a contradiction
Same assistant, same conversation, no restart. Early on it used your name correctly three times. Now it asks who you are. It didn't get worse at listening — something concrete changed about what it could see.
Commit to a prediction
What most likely explains the "forgotten" name?
This lesson changes one thing: you'll stop picturing an AI slowly forgetting like a person, and start picturing a fixed-size desk — everything it can use right now has to physically fit on it, and older pages get pushed off as new ones arrive.
Discover the mechanism
Every inference run (Lesson 9) starts by assembling one bundle of tokens: the provider's hidden instructions, the conversation so far, anything you've pasted in. That bundle has a hard maximum — the context window. It is the only changeable input the machine has; everything else is frozen training (Lesson 5).
Consequences follow directly. A long chat eventually exceeds the budget, so older turns are dropped or compressed — the "forgotten name" mystery solved. A pasted 300-page document may not fit at all, or may fit but be read unevenly (Lesson 8's spotlight doesn't shine equally on a huge desk — items buried in the middle can be under-weighted). And a brand new conversation starts with an empty desk: no carry-over, by default, at all.
Nine times out of ten, when people say an AI "remembers" or "forgets", they are describing desk mechanics, not anything resembling memory.
desk sizes vary enormously between products and change often — treat any specific figure as a snapshot, not a law. Lesson 15 completes the picture: engineered features that genuinely persist information between conversations are a different mechanism again, stocking the desk before you arrive.A brilliant examinee — the trained model — may bring any materials they like. But only what physically fits on the desk in front of them is usable this minute. Slide a new page on, and something else slides off the far edge.
- Desk area = the context window, measured in tokens (Lesson 3).
- Pages on the desk = your messages, its replies, pasted documents, the provider's hidden instructions.
- Sliding off = older turns being truncated or summarised away.
- The examinee's trained skill = the frozen weights (Lesson 5) — knowledge from practice, not from the desk. Training vs context is exactly the exam-hall's skill vs its papers.
Overflow the desk yourself
Conceptual illustration — a deliberately tiny desk (200 tokens) so the limit arrives fast. Feed the chat, watch the gauge, and see what survives.
Ask it:
Where this bites at work
Kerry's care-plan drafting chat degrades after an hour — vaguer suggestions, a repeated question, a detail from early on ignored. She now starts a fresh conversation with a tight summary of the decisions made so far, instead of scrolling back and re-pasting everything. Output improves instantly, and a smaller desk costs less to run (Lesson 9's metering, now a management habit).
Economy here is optimisation, not brevity for its own sake: include everything the task needs, and nothing it doesn't.
A student remembers pages after they slide off a real desk; the model has zero trace of off-desk material — none, ever. And a huge desk isn't used evenly: material buried in the middle of a very long conversation can be under-weighted even while technically "on the desk". Bigger isn't automatically better; well-managed is.
Now make the call
Kerry's hour-long care-plan chat has gone vague and repetitive, and she has ten minutes before her next client. What does she do?
Try asking the assistant to explain its own forgetting too — worth seeing how confidently it answers a question it cannot actually know.
Apply it to your work
Next time a long AI conversation feels like it's losing the plot, don't keep pushing — write a three-line restatement of what matters (decisions made, facts to keep, the current ask) and start fresh with it. Add the habit as a rule on your capstone checklist.
Prove it to yourself
Two quick questions — untimed, retryable, and they count toward ★ Mastered (completed + decision + self-check all correct). This is what defeats the feeling of "that made sense" evaporating by next week.
Explain it in your own words
Two or three sentences for a colleague. Then compare against the rubric — it checks ideas, not wording, and only you see your note.
✓ Lesson 10 stamped: Desk Manager
You overflowed a desk, watched the slide-off, and rescued it with a summary. Lesson 11 asks: now that you know what fits on the desk, how do you word what you put there?
Evidence & review — how we know what this lesson claims
| Claim | Type | Basis | Review risk |
|---|---|---|---|
| Every inference run is bounded by a fixed maximum token budget (the context window) | How it works | Architecture documentation | Low |
| When the budget is exceeded, older content is dropped or compressed rather than the model "forgetting" gradually | How it works | Architecture documentation | Low |
| Material positioned in the middle of very large contexts can be under-weighted ("lost in the middle") | Research finding | Primary research literature | Medium |
| Specific context window sizes and per-product truncation/summarisation behaviour | Product behaviour | Provider documentation | High — reviewed quarterly |
| The desk simulator | Teaching device | Deterministic toy demonstration, labelled illustrative | Medium |
| The open-book exam analogy | Useful mental model | Limits stated in the lesson | Low |
Concept review due: January 2027. Product-behaviour claims: quarterly.
Where you are in the machine
You've stepped onto the scaffolding — the desk everything else in this act gets built on. Select any layer for its role, lessons and your live progress.
The machine, bottom to top: training data (2) → token pieces (3) → prediction engine (1, 4–9) → the desk (10–11) → senses (12) → live knowledge (13) → tools (14) → memory (15) → agent loops (16) → human judgement (17–21).
Next: what you put on the desk
The desk is set. Now — what you put on it, and how you phrase it, changes everything. That craft is prompting.