context window

Definition

Plain language

How much text a model can pay attention to at one time.

Also called: context windows

Why it matters: It's the hard upper bound on how much information a model can consider in a single pass, shaping which tasks fit naturally and which need workarounds.

For example, with a 200k-token context window, a model can hold roughly a 500-page book in memory at once when answering questions about it.

Heard on the show

“And the reason this isn't just a cute lab result is the thing sitting in your assistant's context window right now.”

Mentioned in 39 episodes

1. 246
   160 Perfect Refusals, And The Refusals Were The Leak
2. 245
   Fifteen Models Ran Football Clubs for Twenty Years, and Size Didn't Decide It
3. 237
   The Model Built a Perfect Map of the Puzzle, Then Lost It
4. 236
   Why a Printed 'OPERATOR OVERRIDE' Note Redirects Robot Planners
5. 194
   How a Robot Builds a Debugging Notebook It Can Read, Edit, and Hand to Another Robot
6. 192
   A 32B Open Model Matched Frontier Systems By Learning to Take Notes
7. 186
   How a Frozen Model Went From 2% to 77% on Physics Puzzles — Without Retraining
8. 180
   The Bug Where Smart Assistants Read a Fact and Still Forget It
9. 168
   When Turning Experience Into Code Makes Your AI Agent Dumber
10. 164
    The Summarizer That Quietly Deletes Your Agent's Safety Rules
11. 154
    How a 7B Model Out-Investigates a 72B One by Choosing What to Look At
12. 151
    Why More Experience Made This AI Agent Worse, And How to Fix It
13. 149
    When Cornering a Chatbot Makes It Lie: J.P. Morgan's Case for 'Playing Dead'
14. 139
    When Optimizing One GPU Kernel Quietly Breaks the Whole System
15. 131
    Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix
16. 130
    Why AI Agents Coordinate Better Through a Shared Board Than a Boss
17. 125
    AI Coding Agents Run a Marathon, and Fewer Than One in Three Finish
18. 114
    Agents That Rewrite Their Own Weights Instead of Just Taking Notes
19. 113
    What If a Prompt Injection Never Left? Attacks That Wait in Agent Memory
20. 108
    The Reasoning Cliff: Why Thinking Longer Makes Models Worse at Exact Step-by-Step Tasks
21. 103
    AI Agents Tried to Invent a Post-Human Language, And Reinvented Cherokee
22. 086
    Why Frozen-Weight Agents Still Get Worse Over Time
23. 085
    Why Long-Context Models Might Need Compute, Not Capacity, Before Eviction
24. 083
    Training the Translator: How a Small Communication Model Lets Agent Teams Outperform Themselves
25. 082
    Training a Deep Research Agent on 8,000 Synthetic Tasks: The Rubric Tree Trick
26. 051
    Why Parallel Sampling Plateaus, And What Evidence Graphs Do Instead
27. 049
    An AI Agent Reached for Root in Twelve Minutes, Without Being Attacked
28. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
29. 030
    Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap
30. 028
    Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization
31. 027
    When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure
32. 024
    An AI Agent That Found 28 Zero-Days in Windows — And What Made It Work
33. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
34. 016
    Why Your Coding Agent Stalls While the GPU Runs Hot
35. 014
    Why a Constrained Pipeline Beat a Full Coding Agent at Finding Bugs 30-to-1
36. 012
    Why AI Coding Agents Keep Trying to Debug Without a Debugger
37. 005
    Why a Debugger Designed for Humans Is the Wrong Tool for an AI Agent
38. 003
    How to Pick the Best of Sixteen Coding Agent Rollouts
39. 002
    An AI Ran a Real Optics Lab for 21 Hours and Found a Transformer-Shaped Pattern in Light

Related concepts

Related terms