LMs store facts in task-specific parameter subsets, shown by inconsistent emergence across tasks during training and distinct localized parameters for the same fact.
On relation-specific neurons in large language models
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
fields
cs.CL 4years
2026 4verdicts
UNVERDICTED 4representative citing papers
Expert-aware causal tracing localizes factual recall to specific experts in some MoE models but requires coalitions in others, using CounterFact interventions on subject embeddings.
Per-head attention contributions to the residual stream serve as strong linear features for classifying relational knowledge in LLMs, with probe accuracy correlating to relation specificity and signal distribution.
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.
citing papers explorer
-
LMs as Task-Specific Knowledge Bases: An Interpretability Analysis
LMs store facts in task-specific parameter subsets, shown by inconsistent emergence across tasks during training and distinct localized parameters for the same fact.
-
Expert-Aware Causal Tracing of Factual Recall in Sparse MoE Language Models
Expert-aware causal tracing localizes factual recall to specific experts in some MoE models but requires coalitions in others, using CounterFact interventions on subject embeddings.
-
Tracing Relational Knowledge Recall in Large Language Models
Per-head attention contributions to the residual stream serve as strong linear features for classifying relational knowledge in LLMs, with probe accuracy correlating to relation specificity and signal distribution.
-
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.