Unverified50% confidenceFactExact time
In 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.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThis identification bias is statistically known as Selection Bias, where the sample is determined by observable characteristics rather than being randomly drawn from the overall population.79% similarUnverifiedHuman 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.77% similarUnverifiedIn practice, the anchoring effect of pre-annotation results on humans must be considered: annotators may tend to accept the error types and severity levels given by the model, thereby reducing the independence and diversity of annotation.74% similarUnverifiedWhen the same model is responsible for both generating content and evaluating quality, a self-referential limitation arises where the evaluation criteria align with the model's own generation preferences, forming a self-endorsement loop.74% similarUnverifiedBehavioral label data has inherent defects, conflating candidate exposure opportunities, recruiter subjective preferences, and objective qualifications, making it difficult to isolate the true influence of each factor74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102729API
curl https://kongchang.com/api/v1/knowledge/claims/1102729MCP
get_claim(id=1102729)