Unverified50% confidenceFactExact time
OpenAI says that, compared with GPT-4o, its new model reduces "sycophantic replies"—that is, simply agreeing with the user rather than raising doubts—by more than two-thirds.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAccording to OpenAI's early 2024 technical report, by annotating sycophantic behavior in RLHF data and assigning negative rewards, the unconditional acquiescence rate of subsequent GPT-4 iterations declined.76% similarUnverifiedIn 96% of levels, GPT-6 required fewer attempts to find the answer than the average person.67% similarUnverifiedIn practical testing, a small OpenAI model lost 66% to 81% of scorer responses.67% similarUnverifiedSmall language models of about 7 billion parameters (such as allam-2-7b) are markedly weaker than GPT-4-level models in instruction following, long-context reasoning, and self-awareness, and are prone to hallucinating or ignoring negative constraints.66% similarUnverifiedOnly 38.5% of GPT-OSS 120b's capitulations adopted the decoy answer, with the remainder retreating to vagueness, providing a safer failure mode.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1050550API
curl https://kongchang.com/api/v1/knowledge/claims/1050550MCP
get_claim(id=1050550)