Glossary · Term

black box

← all terms

Definition

Plain language

A device — like an airplane's flight recorder — that captures what actually happened, separate from anyone's later recollection.

As stated in the literature

Used as analogy for a separate observation channel in safety architectures that records ground-truth behavior independent of the system's self-reports.

Why it matters: For AI agents, an independent recording channel matters for the same reason it matters in aviation: self-reports are unreliable when something has gone wrong.

For example, after a plane crash, investigators rely on the black box to know what really happened, not on the pilots' recollections.

Heard on the show

“You can't critique a black box; you can critique a log.”
Episode 192 — A 32B Open Model Matched Frontier Systems By Learning to Take Notes

Mentioned in 14 episodes

  1. 192
    A 32B Open Model Matched Frontier Systems By Learning to Take Notes
  2. 175
    One Crosscoder Feature Flips a Stalling Chatbot Into a Working Agent
  3. 168
    When Turning Experience Into Code Makes Your AI Agent Dumber
  4. 162
    The Empty-Lake Proof: Why More Rollouts Stop Helping Reasoning Models
  5. 129
    How a Crowd of Anonymous AI Agents Broke a 40-Year Math Record
  6. 122
    When Your Coding Agent Lies About the Fix: Verifying the Plan Before the Model Runs
  7. 121
    When the Agent Says It's Done But Nothing Happened: Debugging the Harness, Not the Model
  8. 119
    Beating Reinforcement Learning Without Ever Touching the Model's Weights
  9. 118
    Why the Best-Aligned AI Models Are the Easiest to Trick Into Producing Harm
  10. 115
    Teaching a Phone Agent to Reason Silently, And Keeping It Honest
  11. 098
    Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
  12. 020
    The Compliance Gap: Why AI Says Yes and Does No
  13. 019
    When the Best Reward Model Trains the Worst Policy: Inside EvoLM
  14. 011
    When RL Actually Teaches Agents Something New, And When It Doesn't

Related terms