Unverified50% confidenceOpinionExact time
Transformer models excel at capturing statistical correlations in text, but lack the ability to build world models and cannot internally simulate the causal chains and long-term consequences of physical world operations.
1
Sources
50%
Confidence
Long-term
Relevance
9/1/2026
First Seen
Sources
Related Entities
Related Claims
Unverified大语言模型幻觉是transformer自回归生成的内在属性，模型每步仅根据当前上下文概率分布采样下一个token，不具备对现实世界的接地能力71% similarUnverified2022年论文'Are Transformers Effective for Time Series Forecasting?'发现简单线性模型在多个基准上能击败当时的Transformer变体70% similarUnverifiedDecision Transformer将离线轨迹数据建模为因果序列生成问题，通过引入回报条件使模型能够在不与环境交互的情况下从静态数据集学习条件策略69% similarUnverified在Transformer模型中，追求好的推理（预测行为）与实现完美检索会争夺参数空间，二者相互冲突68% similarUnverified研究表明Transformer模型浅层倾向于捕捉语法和局部模式，深层负责更抽象的语义推理和事实回忆68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/840512API
curl https://kongchang.com/api/v1/knowledge/claims/840512MCP
get_claim(id=840512)