Unverified50% confidenceSolutionExact time
激活修补的操作是对同一模型运行干净输入和污染输入，记录干净运行的中间激活值并注入污染运行，观察哪些替换能恢复正确输出
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
因果理论破解大模型黑箱：机械可解释性的新范式
Related Claims
Unverified消融实验的核心思想是系统性地移除或替换模型中的单个组件，观察整体性能变化，从而量化每个组件的边际贡献69% similarUnverified激活修补由Meng等人在2022年的ROME论文中广泛推广，其原理是将一次运行中某层的激活值移植到另一次运行中观察输出变化，从而进行因果推断64% similarUnverifiedAutoregressive models work by taking the output of the previous step as the input for the next step, generating sequences incrementally.64% similarUnverified重复运行以计算通过率比依赖单次结果更能反映模型的真实表现64% similarVerifiedThe ReAct framework enables the model to interact with the external environment while reasoning—first reasoning about what to do, then calling external tools to act, then observing the results returned by the tools, iterating in a loop until the task is complete.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/509894API
curl https://kongchang.com/api/v1/knowledge/claims/509894MCP
get_claim(id=509894)