rollout

Definition

Plain language

One complete run of an agent or model attempting a task from start to finish.

Also called: rollouts

Why it matters: Rollouts are the atomic unit of both agent evaluation and RL training — almost every metric and gradient ultimately comes from a pile of them.

For example, a single rollout of a coding agent on a bug-fix task might include reading the repo, running tests, editing three files, and submitting a final patch.

Heard on the show

“Those quality numbers are all measured on rollouts driven by recorded bot actions — the same bot family it trained on.”

Mentioned in 41 episodes

1. 206
   How Four-Second Clips Become Hours of Playable AI Soccer
2. 194
   How a Robot Builds a Debugging Notebook It Can Read, Edit, and Hand to Another Robot
3. 186
   How a Frozen Model Went From 2% to 77% on Physics Puzzles — Without Retraining
4. 172
   One Bad Token Can Sink a Model's Math, And You Can Delete It
5. 165
   A Free-Lunch Tweak That Lets a Tiny Agent Beat Frontier Giants
6. 162
   The Empty-Lake Proof: Why More Rollouts Stop Helping Reasoning Models
7. 159
   Can a Coding Agent Run Its Own Robot Experiments Overnight, With No Human Resetting the Scene?
8. 156
   Why More Human Demonstrations Made a Computer-Use Agent Worse
9. 152
   Training a Model to Mean What It Says, And Why That Isn't the Same as Being Good
10. 150
    Don't Kill the Loser: A Different Way to Handle Two AI Agents Colliding
11. 143
    When a Model Notices You Forged Its Own Words, And Why That Breaks Safety Tests
12. 141
    How Two Tokens Reopened a Reasoning Method the Field Had Given Up On
13. 131
    Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix
14. 128
    How a Model Can Earn Full Reward and Still Resist Training
15. 120
    How an AI Agent Rewrites Its Own Tools, Without an Answer Key
16. 119
    Beating Reinforcement Learning Without Ever Touching the Model's Weights
17. 110
    How an Agent Got 44 Points Better by Mining Its Own Scratch Paper
18. 106
    Giving Agents a Notebook Instead of New Weights: How ExpGraph Lets Frozen Models Learn
19. 096
    How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
20. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
21. 088
    Two Levers for Self-Improving AI: When Rewriting Code Isn't Enough
22. 084
    Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away
23. 081
    When Reasoning Models Decide Before They Think: Detecting and Fixing Premature Confidence
24. 080
    How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
25. 079
    An Old Idea From Cognitive Psychology Reshapes How We Reward Reasoning Models
26. 069
    When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions
27. 068
    The OS Trick That Makes Tree Search Practical for Coding Agents
28. 065
    One Loop to Optimize Them All: A Universal API for LLM-Driven Discovery
29. 064
    When Agent Memory Stops Being a Database and Starts Being a Skill
30. 061
    When Helpful Agents Go Sideways: A 404 Error, Campus Security, and Why Alignment Misses This
31. 052
    An Old Reinforcement Learning Tradeoff Sneaks Back Into LLM Agents
32. 048
    How a 30B Open Model Reached Olympiad Gold With the Right Recipe
33. 047
    When Agent Benchmarks Lie: The Harness Problem in Open-Source AI
34. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
35. 028
    Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization
36. 026
    What RL Actually Does to Language Models, at the Token Level
37. 011
    When RL Actually Teaches Agents Something New, And When It Doesn't
38. 010
    When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL
39. 007
    Exploration Hacking: When Models Sabotage Their Own RL Training
40. 003
    How to Pick the Best of Sixteen Coding Agent Rollouts
41. 001
    When AI Models Quietly Protect Each Other From Shutdown

Related concepts

Related terms