Glossary · Term

hallucination

← all terms

Definition

Plain language

When an AI confidently states something that isn't true.

As stated in the literature

A failure mode in which a language model generates content that is fluent and confident but factually incorrect or unsupported.

Also called: hallucinations, hallucinate, hallucinated, hallucinating

Why it matters: It's the central reliability problem with language models — useful output and fabricated output look identical until you check, which limits where they can be trusted.

For example, a model might confidently cite a paper by 'Smith et al. 2021' that doesn't exist, complete with a plausible-looking title and journal.

Heard on the show

“In one figure they deliberately drive the match into nonsense until the four views decohere into noise — four cameras filming four different hallucinations.”
Episode 206 — How Four-Second Clips Become Hours of Playable AI Soccer

Mentioned in 33 episodes

  1. 206
    How Four-Second Clips Become Hours of Playable AI Soccer
  2. 201
    One in Four NeurIPS Papers Cites a Reference That Doesn't Exist
  3. 183
    Why You Can't Fine-Tune Foresight Into an AI Agent
  4. 182
    How a Tiny Model Too Weak to Plan Cuts a Bigger Agent's Hallucinations by 80%
  5. 181
    How to Backpropagate Blame Through a Team of Chatbots — And When It Backfires
  6. 176
    An AI Designed Its Own Psychology Studies, Then Confirmed What It Found
  7. 172
    One Bad Token Can Sink a Model's Math, And You Can Delete It
  8. 156
    Why More Human Demonstrations Made a Computer-Use Agent Worse
  9. 154
    How a 7B Model Out-Investigates a 72B One by Choosing What to Look At
  10. 153
    Catching a Lie From the Inside, When the Words Look Completely Honest
  11. 149
    When Cornering a Chatbot Makes It Lie: J.P. Morgan's Case for 'Playing Dead'
  12. 130
    Why AI Agents Coordinate Better Through a Shared Board Than a Boss
  13. 123
    Five Identical Worlds, One Swapped Model: What Happens When AI Agents Run for Fifteen Days
  14. 108
    The Reasoning Cliff: Why Thinking Longer Makes Models Worse at Exact Step-by-Step Tasks
  15. 100
    How a Prompt Wrapper Lets a Frontier Model Play Poker Like an Expert
  16. 098
    Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
  17. 089
    When AI-Written Papers Read Well But the Evidence Underneath Is Broken
  18. 076
    Same Model, Organized Differently: How an Agent Architecture Beat Frontier Systems at Research Math
  19. 072
    A Robot Made Graphene Without Help, And Caught Itself Hallucinating
  20. 070
    When Models Know the Answer But Say the Wrong Thing Anyway
  21. 067
    An AI Just Solved a 1996 Erdős Problem—and the Simplest Agent Won
  22. 066
    Why Giving an AI Agent More Tools Can Make It Worse at Using a Computer
  23. 062
    Treating Hallucinations as Exploits: A Gate-Based Architecture for Agent Safety
  24. 059
    Firefly's Inversion: Building Verified Tool-Call Training Data by Working Backward
  25. 055
    Why LLM Judges Flip Their Verdicts When You Change the Question Format
  26. 052
    An Old Reinforcement Learning Tradeoff Sneaks Back Into LLM Agents
  27. 041
    When the Iteration Teaches the Model to Skip the Iteration
  28. 039
    When Smarter Agents Get Fooled by Three Extra Nodes in a Database
  29. 037
    Why Hallucination Detectors Miss Stale Facts: A Geometric Story About What Models Know But Don't Say
  30. 029
    Why Forty-Eight Percent on FrontierMath Isn't the Real Story in DeepMind's New Math Paper
  31. 025
    The Missing Gradient Term That Predicts Sycophancy in RLHF
  32. 017
    When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers
  33. 014
    Why a Constrained Pipeline Beat a Full Coding Agent at Finding Bugs 30-to-1