Unverified50% confidenceFactExact time
Selection bias is systematic rather than random: the model may make the same mistake under multiple orderings because of the same underlying preference (such as a tendency toward a particular position or semantic label).
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedInjecting "ties" (i.e., pairs of responses that are equally preferred) into preference data can weaken spurious correlations, because ties provide no discriminative signal and force the model to ignore irrelevant features.71% similarUnverifiedPreference data suffers from selection bias: users willing to make edits are typically only a minority and their editing standards are inconsistent, so filtering, deduplication and consistency checks are usually needed before use.71% similarUnverifiedPointwise scoring reranking is a common paradigm in information retrieval, computing a relevance score for each candidate document independently and then sorting; its drawback is the lack of relative comparison among candidates, making it difficult for the model to calibrate the scale of absolute scores.70% similarUnverifiedEarly research found that even when using examples with randomly incorrect labels, few-shot performance was sometimes not much worse than with correct labels.69% similarUnverifiedWhen multiple algorithms with different inductive biases all converge on the same error pattern, it indicates that the problem lies at the level of problem definition or the feature space rather than the models themselves.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1103096API
curl https://kongchang.com/api/v1/knowledge/claims/1103096MCP
get_claim(id=1103096)