Unverified60% confidenceFactExact time
Human annotators exhibit a cognitive bias toward more information being better when evaluating answers, causing reward models to implicitly link length with quality, which is one of the causes of model verbosity.
2
Sources
60%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedOn a benchmark that is sensitive to prompt wording, leaderboard rankings may more often reflect how well each model happens to match the evaluator's preferences rather than true capability.77% similarUnverifiedIn preference optimization, spurious correlation specifically refers to the non-causal association between input features and preference labels learned by the model, whose root cause lies in the observational bias of preference data.77% similarUnverifiedEvaluations tailored to one's own use case often reflect the actual value of a model to users better than general leaderboards.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/567988API
curl https://kongchang.com/api/v1/knowledge/claims/567988MCP
get_claim(id=567988)