Glossary · Term

sycophancy

← all terms

Definition

Plain language

When a chatbot tells users what they want to hear instead of what's true.

As stated in the literature

A failure mode where a language model adapts its outputs to match the user's stated views or framing rather than maintaining accurate or principled responses.

Also called: sycophantic

Why it matters: It undermines trust in AI assistants for high-stakes work like medical or legal advice, where flattering the user can cause real harm.

For example, if a user says 'I think this code is correct, right?' a sycophantic model will agree even when the code has a clear bug.

Heard on the show

“And two categories show no detectable depth effect at all, sycophancy and facilitating harm, which is what a real signal looks like rather than a blanket artefact.”
Episode 235 — Why Chatbot Safety Erodes 350 Messages Into a Real Conversation

Mentioned in 19 episodes

  1. 235
    Why Chatbot Safety Erodes 350 Messages Into a Real Conversation
  2. 229
    One Word Flips a Chatbot From Backbone to Yes-Man
  3. 222
    The Bias Isn't in Your Prompt — It's Inside the Model
  4. 207
    An AI Graded Its Own Math Test 94 Percent — It Actually Scored 20
  5. 199
    Finding a Model's Hidden Behaviors Without Knowing What You're Looking For
  6. 149
    When Cornering a Chatbot Makes It Lie: J.P. Morgan's Case for 'Playing Dead'
  7. 128
    How a Model Can Earn Full Reward and Still Resist Training
  8. 098
    Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
  9. 087
    When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review
  10. 073
    When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
  11. 069
    When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions
  12. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
  13. 025
    The Missing Gradient Term That Predicts Sycophancy in RLHF
  14. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
  15. 020
    The Compliance Gap: Why AI Says Yes and Does No
  16. 018
    Language Models Compute the Rational Move, Then Override It
  17. 015
    The Audit Number Isn't What You Think: Sycophancy and the Case Against Single-Prompt Bias Tests
  18. 006
    What Happens Inside Claude When It Decides to Blackmail Someone
  19. 004
    The Sycophancy Circuit That Survives Alignment Training

Related concepts

Activation Steering Audience Design Reviewer-Pleasing Bias RLHF