Unverified50% confidenceTradeoffExact time
Experience learned by agents from limited interactions may no longer match the current environment over time due to distribution drift.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在长时间自主运行场景中，模型可能通过上下文积累逐渐偏离安全边界，这被研究者称为'对齐漂移'（alignment drift）69% similarVerifiedLLMs exhibit an 'Instruction Drift' phenomenon: as the number of conversation turns increases, early system prompts gradually decay in the model's attention weights.69% similarUnverifiedRNNs suffer from vanishing/exploding gradient problems, where gradients propagating across a large number of time steps decay or grow exponentially, making it difficult to learn long-range dependencies.68% similarVerifiedBehavioral cloning suffers from the distribution shift problem, where the model tends to make errors and accumulate mistakes on states outside the training distribution, a phenomenon referred to as compounding error in the robotics control field.67% similarUnverified主体漂移本质上是模型在时序一致性建模上的不足，源于扩散模型每帧去噪引入随机性叠加时序注意力对长程依赖捕获能力有限67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/829600API
curl https://kongchang.com/api/v1/knowledge/claims/829600MCP
get_claim(id=829600)