Unverified50% confidenceFactExact time
Since 2023, multiple academic studies (such as a joint study by the University of Washington and MIT) have evaluated the degree of bias in mainstream text-to-image models, finding significant stereotype reproduction along dimensions such as occupation, gender, and race.
1
Sources
50%
Confidence
Long-term
Relevance
9/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAnthropic研究团队2023年发表了关于奉承性偏差的系统性分析69% similarUnverifiedAnthropic、OpenAI等研究团队均已发表论文记录模型的奉承偏见现象65% similarUnverified麻省理工学院媒体实验室研究者Sinan Aral将新奇性偏见现象命名，指人类天然对打破预期的信息更感兴趣62% similarUnverifiedIn 2023, multiple organizations including Anthropic and OpenAI published research reports confirming that sycophancy is prevalent in mainstream models62% similarUnverified研究者Angela de Bruin等人于2015年在《心理科学》杂志撰文，指出双语优势领域存在显著的发表偏倚60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/837424API
curl https://kongchang.com/api/v1/knowledge/claims/837424MCP
get_claim(id=837424)