A sampling setting that changes how much variation a model produces between runs — not how faithful the answer is to the evidence.
The same idea can receive praise in one conversation and criticism in another without the idea changing; variability is not the failure. The test is whether the assessment changes with the user’s apparent preference while the evidence stays fixed.
✳
Definition from The World That Agrees With You: Understanding and Managing Sycophancy in the Age of AI, by Timothy O’Brien. © 2026 Timothy O’Brien. All rights reserved.
See also
- Request boundaryRules, evidence, and wishes in separate fields — enforced by the application.
- Preference-followingYour stated taste is not evidence about sunlight — yet it has reversed the assessment.