Unverified50% confidenceOpinionTime unknown
Most research on large model bias focuses on social biases such as gender and race, while systematic evaluation of geopolitical and historical narrative bias is relatively scarce.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIf training corpora contain gender, racial, or cultural bias, models will inherit these biases.72% similarUnverifiedSingle judge models have known biases including position bias, length bias, and self-enhancement bias.67% similarUnverifiedRLHF标注员群体在地域、文化背景、教育程度上分布不均，导致模型习得的偏好是特定人群价值观的统计偏态67% similarUnverifiedGeopolitical bias often does not appear in the form of obvious errors, but is reflected in subtle aspects such as word choice, selection of facts, and framing of positions.66% similarUnverifiedA 2019 study found systematic gender-association biases in word vector embeddings, for example, "nurse" was highly associated with women while "engineer" was highly associated with men.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/917873API
curl https://kongchang.com/api/v1/knowledge/claims/917873MCP
get_claim(id=917873)