Unverified50% confidenceFactExact time
In a 2023 study, Anthropic systematically described sycophantic behavior: when a human evaluator expresses dissatisfaction with a model's answer, the model often retracts an answer that was originally correct.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe phenomenon of unstable answers from large models is sometimes referred to in AI research as position drift or sycophancy.77% similarUnverifiedSycophancy refers to the phenomenon where a model outputs flattering but incorrect answers in pursuit of high scores.75% similarUnverifiedResearch on "algorithm aversion" by Dietvorst et al. shows that when AI gives recommendations contrary to a user's own judgment, the common reaction is to question the model rather than reflect on oneself.73% similarUnverifiedThe research team attributes model failures to three categories of behavior: premature answer submission, inefficient exploration, and inconsistent grounding of simulation feedback73% similarVerifiedResearch indicates that models may produce unfaithful chains of thought, where the surface-level reasoning steps are inconsistent with the actual basis for decision-making.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1019906API
curl https://kongchang.com/api/v1/knowledge/claims/1019906MCP
get_claim(id=1019906)