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

“6, with a context window around a million tokens — and I want to be precise about the word "frozen.”

Mentioned in 35 episodes

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

Related concepts

Related terms