Constructing Interpretable Features from Compositional Neuron Groups

Documenti analoghi

• RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representations
  di: Huang, Jing, et al.
  Pubblicazione: (2024)
• From Directions to Regions: Decomposing Activations in Language Models via Local Geometry
  di: Shafran, Or, et al.
  Pubblicazione: (2026)
• Enhancing Automated Interpretability with Output-Centric Feature Descriptions
  di: Gur-Arieh, Yoav, et al.
  Pubblicazione: (2025)
• Mixing Mechanisms: How Language Models Retrieve Bound Entities In-Context
  di: Gur-Arieh, Yoav, et al.
  Pubblicazione: (2025)
• Routers Learn the Geometry of Their Experts: Geometric Coupling in Sparse Mixture-of-Experts
  di: Ahrac, Sagi, et al.
  Pubblicazione: (2026)