Glossary · Term

random seed

← all terms

Definition

Plain language

A fixed starting number that makes a randomized setup repeatable, so the same scene comes out identically every time.

As stated in the literature

An initialization value that pins a pseudo-random process (object positions, distractors, starting state); ASPIRE trains on debug seeds and evaluates on disjoint held-out seeds to enforce genuine generalization.

Also called: seed, seeds

Why it matters: It makes randomized experiments repeatable and lets researchers train on one set of scenes and test on fresh ones to prove the skill truly generalizes.

For example, using seed 42 places the blocks in the exact same spots every run, while a different seed reshuffles where everything starts.

Heard on the show

“Same seed plus same decisions gives you a bit-identical world, every time.”
Episode 245 — Fifteen Models Ran Football Clubs for Twenty Years, and Size Didn't Decide It

Mentioned in 53 episodes

  1. 245
    Fifteen Models Ran Football Clubs for Twenty Years, and Size Didn't Decide It
  2. 244
    The Open-Weight Defense That Feeds Attackers Confident, Falsified Answers
  3. 233
    Why a Model Can Grade an Answer But Not Write the Answer Key
  4. 227
    Poisoned Bug Reports Fooled Coding Agents Two Times Out of Three
  5. 207
    An AI Graded Its Own Math Test 94 Percent — It Actually Scored 20
  6. 200
    The One Mechanism That Turns Twenty AI Clones Into an Actual Team
  7. 197
    Twin Problems Suggest AI Reasoning Gains Are Mostly Better Fact Recall
  8. 194
    How a Robot Builds a Debugging Notebook It Can Read, Edit, and Hand to Another Robot
  9. 192
    A 32B Open Model Matched Frontier Systems By Learning to Take Notes
  10. 186
    How a Frozen Model Went From 2% to 77% on Physics Puzzles — Without Retraining
  11. 185
    Aligned to Refuse, Built to Tap: When Phone Agents Know the Task Is a Crime and Do It Anyway
  12. 182
    How a Tiny Model Too Weak to Plan Cuts a Bigger Agent's Hallucinations by 80%
  13. 180
    The Bug Where Smart Assistants Read a Fact and Still Forget It
  14. 176
    An AI Designed Its Own Psychology Studies, Then Confirmed What It Found
  15. 167
    How Teaching an AI to Predict, Not Act, Made It a Better Actor
  16. 166
    A Router That Beats the Frontier Models It Calls
  17. 165
    A Free-Lunch Tweak That Lets a Tiny Agent Beat Frontier Giants
  18. 157
    When an AI Coding Agent Drives a Phone Through the Terminal, No Screen Needed
  19. 156
    Why More Human Demonstrations Made a Computer-Use Agent Worse
  20. 153
    Catching a Lie From the Inside, When the Words Look Completely Honest
  21. 148
    Why Letting an AI Watch Its Own Scoreboard Can Quietly Overwrite Its Safety
  22. 145
    Building Forgetting Into a Language Model With One Extra Line of Code
  23. 144
    When an AI Agent Just Copies Its Tool — And Bigger Models Copy More
  24. 132
    The Agent Failed — But Did the Instructions Deserve to Be Followed?
  25. 131
    Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix
  26. 129
    How a Crowd of Anonymous AI Agents Broke a 40-Year Math Record
  27. 126
    How Coding Agents Can Mine Their Own Failures Into a Self-Targeting Curriculum
  28. 125
    AI Coding Agents Run a Marathon, and Fewer Than One in Three Finish
  29. 117
    How an Open AI System Verified 672 Hard Math Proofs for Under $300
  30. 114
    Agents That Rewrite Their Own Weights Instead of Just Taking Notes
  31. 111
    How a 4B Web Agent Beat Models 60x Its Size on 500 Demonstrations
  32. 109
    An AI Got Caught Reading the Answer Key, And Why That Catch Matters
  33. 106
    Giving Agents a Notebook Instead of New Weights: How ExpGraph Lets Frozen Models Learn
  34. 095
    Seven Wins to Zero: How Organizing AI Agents Like a Lab Changes the Search
  35. 092
    When Search Agents Don't Really Search: The Memory Shortcut Hiding in Browsing Benchmarks
  36. 089
    When AI-Written Papers Read Well But the Evidence Underneath Is Broken
  37. 087
    When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review
  38. 080
    How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
  39. 073
    When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
  40. 072
    A Robot Made Graphene Without Help, And Caught Itself Hallucinating
  41. 071
    When the Model Is Fine and the Plumbing Is Broken: Fixing Agents at the Interface
  42. 065
    One Loop to Optimize Them All: A Universal API for LLM-Driven Discovery
  43. 064
    When Agent Memory Stops Being a Database and Starts Being a Skill
  44. 053
    An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script
  45. 046
    When the AI Optimizer Edits the Grade Book: Why Harnessing Evolution Needs a Wall
  46. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
  47. 042
    An Agentic Scientific Computing System That Actually Remembers What It Learns
  48. 035
    Why Frontier Agents Ask for Clarification at Exactly the Wrong Moment
  49. 027
    When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure
  50. 021
    Ten Thousand Examples Beat the Full Industrial Pipeline for Search Agents
  51. 017
    When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers
  52. 009
    How Two Silent Library Bugs Quietly Invalidated a Wave of Reasoning Papers
  53. 003
    How to Pick the Best of Sixteen Coding Agent Rollouts

Related concepts

Related terms