Context rot: when more context hurts
The model reads everything you give it, so just paste in all your docs and let it sort them out, right?
The idea inside
Past a point, extra context backfires: the key fact gets buried and answers slip.
After this lesson
You can explain context rot: why more context can lower answer quality, and the habits (curate, compress, retrieve, cite, verify) that beat dumping everything in.
Where it leads
Deciding what the model should see is real work, next, the tools and loops that manage it for you.
Inside this lesson
That's the real lesson stage, paused. Claim your pass to operate it.
See how AI actually works, end to end.
This lesson is one stop on the full arc. Unlock all of it, and keep it for life.
What you get
- The 34-lesson main path, a finishable route from a word to agents
- Goal tracks for using AI at work and building AI features
- Boss labs that make you apply a whole act, not just recognize it
- Spaced recall that brings each idea back before you forget
- Course memory: every term defined, with links to where it first appears
- A shareable capability card when you finish the main path
- Lifetime access on every device, every future lesson included
Not videos to watch. You predict, operate the machine, then prove it. That is why it stays.
99 interactive lessons and challenges. No videos, no code.
Free launch pass: lifetime access, no card needed
New here? The first lessons are free to try. Start with lesson 0.1
What this lesson shows
Past a point, extra context backfires: the key fact gets buried and answers slip.
The question it opens with
The model reads everything you give it, so just paste in all your docs and let it sort them out, right?
The walkthrough, in the lesson's own words
- See it? Burying the needle drags accuracy down.
- One needle, no filler. Drag the slider a notch and watch the accuracy gauge.
- Keep padding. Watch accuracy sag as the needle gets buried.
- Drag the slider to pad the prompt with irrelevant text.
- Signal beats volume. Give it the few right chunks, not everything.
- The gauge shows a real, documented effect, not measured numbers.
- Past a point, the extra text buries the key fact (a needle in a haystack), spreads attention (lesson 3.3) thinner across more words, and burns more of the fixed budget (lessons 4.2 / 6.2). Accuracy tends to drop, and it costs more and runs slower.
- Same model, same needle, same question, only the pile of irrelevant text changed. The answer got harder to find while time and cost went up.
- The model reads everything, so more context is always better. Just paste in all your docs and let it sort them out.
- Past a point, extra text buries the key fact and spreads attention thinner, so accuracy drops while cost and latency rise. A few relevant chunks beat a giant pile.
- You paste a whole 50-page manual into a chatbot and ask one specific question, but the answer it gives is worse than when you pasted just the relevant page. Why?
- Context rot: past a point, more text buries the key fact and spreads attention thinner, so accuracy drops while cost and latency rise. Feed it the few chunks that matter, not everything you have.
- The key fact is buried among many lines of irrelevant filler text.
Key takeaway
You watched accuracy fall as you padded the prompt with filler, signal beats volume.
What you can do after this lesson
You can explain context rot: why more context can lower answer quality, and the habits (curate, compress, retrieve, cite, verify) that beat dumping everything in.
Check yourself: You paste more and more into the prompt and the answers get worse, not better. Why?
- Past a point the key fact gets buried in a huge window and accuracy slips (context rot)(correct)
- The model runs out of disk space
- More context always helps, so something else must be wrong
- The temperature rose on its own
A bigger window isn't free. As it fills, the model skims and the key detail gets lost in the middle. Trimming to what matters often beats pasting everything.
Where it leads: Deciding what the model should see is real work, next, the tools and loops that manage it for you.
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.