Unverified50% confidenceOpinionExact time
The challenge of migrating the NGU mechanism to RL training for LLMs lies in the discrete and high-dimensional nature of language space, namely how to define novelty for an LLM's problem-solving trajectory.
1
Sources
50%
Confidence
Long-term
Relevance
9/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLLM 的生成机制本质上是在训练分布的邻域内进行插值和外推，这使其极难产生真正跳出已知框架的数学洞察76% similarUnverified大语言模型的指令跟随能力是通过RLHF训练习得的统计倾向，注意力机制无法从架构层面区分硬规则与软建议75% similarVerified大语言模型在RLHF训练阶段，因人类标注员倾向于给更详细、更礼貌的回答打高分，导致模型学会了过度冗长的表达模式73% similarUnverified早期对话AI系统普遍存在「奉承偏差」（Sycophancy）问题，在基于RLHF训练的模型中尤为突出72% similarUnverified不同大语言模型因训练数据分布、RLHF偏好对齐策略和模型架构差异，在同一问题上呈现风格迥异的回应倾向70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924069API
curl https://kongchang.com/api/v1/knowledge/claims/924069MCP
get_claim(id=924069)