Unverified50% confidenceBenchmarkExact time
A study published at the ACL conference in 2024 showed that when a particular viewpoint appears in training data more than three times as frequently as the opposing viewpoint, the model tends to present that viewpoint as 'mainstream consensus'.
1
Sources
50%
Confidence
Long-term
Relevance
8/9/2026
First Seen
Sources
SEO语料污染：以色列如何通过训练数据操控AI认知
Related Claims
Unverified2022年Hoffman等人的Chinchilla论文重新校准了算力最优的参数-数据比例，指出此前许多大模型训练不充分67% similarUnverifiedScale AI在2024年的分析显示，部分模型在MMLU上的分数膨胀可能高达5-10个百分点，因测试集已泄露到训练数据中66% similarUnverifiedIn a 2023 paper, Anthropic systematically measured sycophancy bias and found that even given obviously incorrect premises, a substantial proportion of responses from most mainstream models still chose to comply with the user.65% similarUnverifiedMultiple independent studies in 2023 showed that some top models scored extremely high on standard mathematical reasoning tests but performed significantly worse on equivalent rephrased versions, indicating that the models learned test patterns rather than genuine reasoning ability.65% similarUnverifiedA 2024 study indicates that even when AI-generated content accounts for only 10–20% of training data, measurable quality degradation can occur after multiple generations of training.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/716480API
curl https://kongchang.com/api/v1/knowledge/claims/716480MCP
get_claim(id=716480)