Glossary · Term

calibration

← all terms

Definition

Plain language

Whether a model's confidence actually matches how often it's right — a well-calibrated model is sure when it should be sure and unsure when it shouldn't.

As stated in the literature

The degree to which a model's predicted probabilities match empirical outcome frequencies, distinct from raw accuracy; instruction tuning and RLHF often degrade it, yielding confident-but-wrong outputs, and conformal methods restore it as a controllable target rate.

Also called: calibrated, well-calibrated, miscalibration, calibration loss

Why it matters: Without it a model can sound completely confident while being wrong, which is dangerous when people trust its answers for medical, legal, or financial decisions.

For example, a well-calibrated weather model that says '70% chance of rain' should actually see rain on about 70 out of 100 such days.

Heard on the show

“… search results might be ambiguous, propose a conditional follow-up if they are — and it gives a calibrated ninety percent instead of a glib hundred. …”
Episode 183 — Why You Can't Fine-Tune Foresight Into an AI Agent

Mentioned in 37 episodes

  1. 183
    Why You Can't Fine-Tune Foresight Into an AI Agent
  2. 182
    How a Tiny Model Too Weak to Plan Cuts a Bigger Agent's Hallucinations by 80%
  3. 179
    How DeepSeek Made One User Faster Without Slowing Down the Crowd
  4. 178
    How an AI Reviewer Learned to Stop Going Easy on AI Writing
  5. 170
    When a One-Liner Beats Your Agent's Clever Verification Logic
  6. 162
    The Empty-Lake Proof: Why More Rollouts Stop Helping Reasoning Models
  7. 139
    When Optimizing One GPU Kernel Quietly Breaks the Whole System
  8. 132
    The Agent Failed — But Did the Instructions Deserve to Be Followed?
  9. 120
    How an AI Agent Rewrites Its Own Tools, Without an Answer Key
  10. 097
    Same Tokens, Same Cost, Wildly Different Results: What Actually Scales in AI Agents
  11. 095
    Seven Wins to Zero: How Organizing AI Agents Like a Lab Changes the Search
  12. 093
    A Calibrated Knob for Weak-to-Strong AI Oversight, Tested on Real Code
  13. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
  14. 080
    How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
  15. 077
    Reading a Model's Confidence Curve to Decide When Chain-of-Thought Is Worth It
  16. 070
    When Models Know the Answer But Say the Wrong Thing Anyway
  17. 069
    When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions
  18. 063
    Why Web Agents Are Slow: A Compiler-Style Fix for Computer-Use Latency
  19. 059
    Firefly's Inversion: Building Verified Tool-Call Training Data by Working Backward
  20. 058
    Why Upgrading Your AI Auditor to a Smarter Model Can Make Your System Less Safe
  21. 053
    An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script
  22. 051
    Why Parallel Sampling Plateaus, And What Evidence Graphs Do Instead
  23. 047
    When Agent Benchmarks Lie: The Harness Problem in Open-Source AI
  24. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
  25. 040
    Two Frozen Models Learn to Whisper: Coupling Through Hidden States
  26. 037
    Why Hallucination Detectors Miss Stale Facts: A Geometric Story About What Models Know But Don't Say
  27. 035
    Why Frontier Agents Ask for Clarification at Exactly the Wrong Moment
  28. 034
    Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool
  29. 026
    What RL Actually Does to Language Models, at the Token Level
  30. 025
    The Missing Gradient Term That Predicts Sycophancy in RLHF
  31. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
  32. 020
    The Compliance Gap: Why AI Says Yes and Does No
  33. 013
    Why Search Keeps Rediscovering the Same Workflow, and What That Means
  34. 010
    When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL
  35. 009
    How Two Silent Library Bugs Quietly Invalidated a Wave of Reasoning Papers
  36. 008
    Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps
  37. 007
    Exploration Hacking: When Models Sabotage Their Own RL Training

Related concepts

Related terms