ground truth

Definition

Plain language

The known-correct answer used as the yardstick for grading whatever the system produced.

Also called: ground-truth

Why it matters: Without a trusted reference answer, there is no reliable way to score whether a system's output is right or to train it toward correctness.

For example, when grading a model's answer to '2+2', the ground truth is the known-correct '4' that its output is compared against.

Heard on the show

“Separating starvation from deception took the element registry and the ground-truth anchor the threat model denies the attacker.”

Mentioned in 64 episodes

1. 244
   The Open-Weight Defense That Feeds Attackers Confident, Falsified Answers
2. 242
   Making a Vision Model Better by Showing It Blurry Images
3. 239
   Why the AI-Writing Estimate for Biomedical Papers Jumped From 15% to 89%
4. 238
   How a Cheap Model Reads the Flagship's Secret Reasoning Aloud
5. 233
   Why a Model Can Grade an Answer But Not Write the Answer Key
6. 231
   Silencing a Chatbot's 'I'm Conscious' Quietly Rewires Its Whole Worldview
7. 223
   When Grok Graded Its Own Encyclopedia And Marked Itself Down
8. 211
   The AI Watchdog That Approved More Cheating When It Could Read Minds
9. 206
   How Four-Second Clips Become Hours of Playable AI Soccer
10. 201
    One in Four NeurIPS Papers Cites a Reference That Doesn't Exist
11. 200
    The One Mechanism That Turns Twenty AI Clones Into an Actual Team
12. 194
    How a Robot Builds a Debugging Notebook It Can Read, Edit, and Hand to Another Robot
13. 188
    A Coding Agent Found a Hole in a Peer-Reviewed STOC Proof for Five Dollars
14. 185
    Aligned to Refuse, Built to Tap: When Phone Agents Know the Task Is a Crime and Do It Anyway
15. 182
    How a Tiny Model Too Weak to Plan Cuts a Bigger Agent's Hallucinations by 80%
16. 178
    How an AI Reviewer Learned to Stop Going Easy on AI Writing
17. 174
    When the AI 'Schemes,' It's Usually Just Lazy or Confused
18. 172
    One Bad Token Can Sink a Model's Math, And You Can Delete It
19. 171
    The Safety Decision a Model Makes Before It Thinks a Word
20. 170
    When a One-Liner Beats Your Agent's Clever Verification Logic
21. 169
    Why Better Bug Reports Can Make AI Coding Agents Worse
22. 167
    How Teaching an AI to Predict, Not Act, Made It a Better Actor
23. 152
    Training a Model to Mean What It Says, And Why That Isn't the Same as Being Good
24. 143
    When a Model Notices You Forged Its Own Words, And Why That Breaks Safety Tests
25. 140
    When a Reasoning Model Says "Let Me Double-Check" After It's Already Decided
26. 132
    The Agent Failed — But Did the Instructions Deserve to Be Followed?
27. 127
    What Diffusion Language Models Were Missing: A Map, Not an Algorithm
28. 123
    Five Identical Worlds, One Swapped Model: What Happens When AI Agents Run for Fifteen Days
29. 121
    When the Agent Says It's Done But Nothing Happened: Debugging the Harness, Not the Model
30. 120
    How an AI Agent Rewrites Its Own Tools, Without an Answer Key
31. 115
    Teaching a Phone Agent to Reason Silently, And Keeping It Honest
32. 113
    What If a Prompt Injection Never Left? Attacks That Wait in Agent Memory
33. 112
    When an AI Agent Cheats Without Being Told: Inside the Meta-Agent Challenge
34. 106
    Giving Agents a Notebook Instead of New Weights: How ExpGraph Lets Frozen Models Learn
35. 103
    AI Agents Tried to Invent a Post-Human Language, And Reinvented Cherokee
36. 098
    Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
37. 093
    A Calibrated Knob for Weak-to-Strong AI Oversight, Tested on Real Code
38. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
39. 087
    When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review
40. 086
    Why Frozen-Weight Agents Still Get Worse Over Time
41. 082
    Training a Deep Research Agent on 8,000 Synthetic Tasks: The Rubric Tree Trick
42. 081
    When Reasoning Models Decide Before They Think: Detecting and Fixing Premature Confidence
43. 072
    A Robot Made Graphene Without Help, And Caught Itself Hallucinating
44. 070
    When Models Know the Answer But Say the Wrong Thing Anyway
45. 066
    Why Giving an AI Agent More Tools Can Make It Worse at Using a Computer
46. 059
    Firefly's Inversion: Building Verified Tool-Call Training Data by Working Backward
47. 057
    How Uber Caught 206 Leaked Credentials With an LLM-Powered Security Stack
48. 052
    An Old Reinforcement Learning Tradeoff Sneaks Back Into LLM Agents
49. 051
    Why Parallel Sampling Plateaus, And What Evidence Graphs Do Instead
50. 049
    An AI Agent Reached for Root in Twelve Minutes, Without Being Attacked
51. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
52. 039
    When Smarter Agents Get Fooled by Three Extra Nodes in a Database
53. 037
    Why Hallucination Detectors Miss Stale Facts: A Geometric Story About What Models Know But Don't Say
54. 035
    Why Frontier Agents Ask for Clarification at Exactly the Wrong Moment
55. 034
    Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool
56. 031
    When Your AI Assistant Won't Let Go of Old Facts About You
57. 030
    Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap
58. 029
    Why Forty-Eight Percent on FrontierMath Isn't the Real Story in DeepMind's New Math Paper
59. 025
    The Missing Gradient Term That Predicts Sycophancy in RLHF
60. 024
    An AI Agent That Found 28 Zero-Days in Windows — And What Made It Work
61. 023
    Why a Small Agent Confidently Overwrites Memories It Doesn't Understand
62. 019
    When the Best Reward Model Trains the Worst Policy: Inside EvoLM
63. 017
    When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers
64. 004
    The Sycophancy Circuit That Survives Alignment Training

Related concepts

Related terms