Do Sparse Autoencoders Identify Reasoning Features in Language Models?

Documenti analoghi

• Spooky Action at a Distance: Normalization Layers Enable Side-Channel Spatial Communication
  di: Pfrommer, Samuel, et al.
  Pubblicazione: (2025)
• Revealing Treatment Non-Adherence Bias in Clinical Machine Learning Using Large Language Models
  di: Liang, Zhongyuan, et al.
  Pubblicazione: (2025)
• Reinforcement Learning for Flow-Matching Policies
  di: Pfrommer, Samuel, et al.
  Pubblicazione: (2025)
• Absence of spurious solutions far from ground truth: A low-rank analysis with high-order losses
  di: Ma, Ziye, et al.
  Pubblicazione: (2024)
• Transport of Algebraic Structure to Latent Embeddings
  di: Pfrommer, Samuel, et al.
  Pubblicazione: (2024)