linear probe

Definition

Plain language

A small simple classifier trained on a model's internal states to test what information they contain.

Also called: linear probing, linear classifier, linear probes, probe, probes

Why it matters: Linear probes are a cheap, standard way to ask 'is this concept actually represented here?' without invasive interventions.

For example, training a one-layer classifier on a transformer's middle layer can reveal whether the model already 'knows' the part of speech of each word at that depth.

Heard on the show

“Second, the linear probe, which reads them in the weakest possible way: multiply by a fixed set of weights, add them up, and output a guess.”

Mentioned in 44 episodes

1. 204
   The Length Estimate Hiding Inside a Word-by-Word Model
2. 191
   How One Researcher Beat GPT-5.2 and Gemini 3 by Judging Their Answers, Not Improving Them
3. 177
   Why Raw Profiler Data Made an AI Worse at Writing GPU Code
4. 171
   The Safety Decision a Model Makes Before It Thinks a Word
5. 164
   The Summarizer That Quietly Deletes Your Agent's Safety Rules
6. 158
   How Floating-Point Rounding Lets a Model Tell Which Chip It's On — And Misbehave
7. 157
   When an AI Coding Agent Drives a Phone Through the Terminal, No Screen Needed
8. 152
   Training a Model to Mean What It Says, And Why That Isn't the Same as Being Good
9. 148
   Why Letting an AI Watch Its Own Scoreboard Can Quietly Overwrite Its Safety
10. 141
    How Two Tokens Reopened a Reasoning Method the Field Had Given Up On
11. 140
    When a Reasoning Model Says "Let Me Double-Check" After It's Already Decided
12. 131
    Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix
13. 125
    AI Coding Agents Run a Marathon, and Fewer Than One in Three Finish
14. 115
    Teaching a Phone Agent to Reason Silently, And Keeping It Honest
15. 100
    How a Prompt Wrapper Lets a Frontier Model Play Poker Like an Expert
16. 094
    Chain-of-Thought Monitoring Fails Across Languages, and Worst Where It's Needed Most
17. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
18. 087
    When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review
19. 086
    Why Frozen-Weight Agents Still Get Worse Over Time
20. 084
    Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away
21. 081
    When Reasoning Models Decide Before They Think: Detecting and Fixing Premature Confidence
22. 077
    Reading a Model's Confidence Curve to Decide When Chain-of-Thought Is Worth It
23. 073
    When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
24. 070
    When Models Know the Answer But Say the Wrong Thing Anyway
25. 069
    When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions
26. 061
    When Helpful Agents Go Sideways: A 404 Error, Campus Security, and Why Alignment Misses This
27. 055
    Why LLM Judges Flip Their Verdicts When You Change the Question Format
28. 054
    When Models Learn the Monitor Exists, the Reasoning Trace Stops Being a Window
29. 051
    Why Parallel Sampling Plateaus, And What Evidence Graphs Do Instead
30. 043
    When 'This Is False' Doesn't Stick: Why Models Learn the Lie Anyway
31. 037
    Why Hallucination Detectors Miss Stale Facts: A Geometric Story About What Models Know But Don't Say
32. 032
    A Sticky-Note for Every Layer: Letting Transformers Remember What They Were Just Thinking
33. 031
    When Your AI Assistant Won't Let Go of Old Facts About You
34. 030
    Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap
35. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
36. 018
    Language Models Compute the Rational Move, Then Override It
37. 015
    The Audit Number Isn't What You Think: Sycophancy and the Case Against Single-Prompt Bias Tests
38. 012
    Why AI Coding Agents Keep Trying to Debug Without a Debugger
39. 011
    When RL Actually Teaches Agents Something New, And When It Doesn't
40. 007
    Exploration Hacking: When Models Sabotage Their Own RL Training
41. 006
    What Happens Inside Claude When It Decides to Blackmail Someone
42. 005
    Why a Debugger Designed for Humans Is the Wrong Tool for an AI Agent
43. 004
    The Sycophancy Circuit That Survives Alignment Training
44. 002
    An AI Ran a Real Optics Lab for 21 Hours and Found a Transformer-Shaped Pattern in Light

Related concepts

Related terms