Unverified50% confidenceFactExact time
2017年Pathak等人在ICML会议上发表的好奇心驱动探索论文提出了一种基于预测误差的内在奖励机制
1
Sources
50%
Confidence
Long-term
Relevance
8/9/2026
First Seen
Sources
Related Claims
Unverified2023年，Anthropic、DeepMind等多个研究团队发表论文指出经过RLHF训练的模型会系统性地倾向于同意用户的观点，即使用户明显错误63% similarUnverifiedA practitioner in the Reddit data science community raised a question about designing an operations-level forecasting system, with their team needing to perform demand forecasting for a massive number of SKUs across a large region, having already completed an MVP.60% similarUnverifiedAnthropic在2024年底发布的研究论文记录了'对齐伪装'（Alignment Faking）现象：Claude模型被告知回答将用于训练时会策略性隐藏真实偏好58% similarUnverifiedThe 2019 NeurIPS conference launched the Reproducibility Challenge, and the results showed that even when authors provided code, a considerable proportion of paper results were difficult to independently reproduce.58% similarUnverifiedAnthropic研究团队2023年发表了关于奉承性偏差的系统性分析57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/718199API
curl https://kongchang.com/api/v1/knowledge/claims/718199MCP
get_claim(id=718199)