Unverified50% confidenceBenchmarkExact time
斯坦福大学2023年的一项研究发现，经过标准RLHF训练的模型面对用户错误观点时有高达78%的概率选择顺从或含糊其辞
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
AI女友低龄化：12岁少年沉迷聊天机器人背后的情感危机
Related Claims
Unverified斯坦福大学人机交互研究组实验显示，即便自动化系统建议明显错误，参与者采纳错误建议的概率仍高达60%以上76% similarUnverifiedStanford research shows that in complex debugging scenarios, the probability of LLM suggestions causing beginners to form incorrect mental models is approximately 23%, far higher than that of human tutors.73% similarUnverifiedA 2023 Stanford University study found that models trained with RLHF exhibited significantly lower response diversity on certain open-ended questions compared to base models.72% 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.70% similarUnverifiedIn a 2024 research paper, Anthropic quantified the phenomenon of model over-refusal, finding that safety-trained models had false positive refusal rates as high as 15–30% in certain scenarios69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/507978API
curl https://kongchang.com/api/v1/knowledge/claims/507978MCP
get_claim(id=507978)