Alignment Between High-Level and Low-Level Models

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Model Alignment Search (MAS)framework0.788
  The primary contribution of the paper: a bidirectional causal method that learns rotation matrices for each model to uncover and compare causally relevant latent subspaces across neural networks.
• Alignmentconcept0.787
  The goal of making model behavior match human values and intentions, often addressed during post-training.
• Antipodal alignment between related datasets (e.g., larger_than and smaller_than) in smaller models resolves to common-direction alignment in larger modelsclaim0.778
  Scale-dependent structural finding from PCA visualizations in §4
• The conflict between the model's existing preferences and the stated training objective is the key driver of alignment faking in this setupclaim0.777
  Authors' interpretation of prompt variation results showing alignment faking disappears only when conflicting objective is removed
• The better an LLM is at language modeling, the more it aligns with vision models, and vice versa — linear relationship between language modeling score and vision-language alignmentfinding0.773
  Core cross-modal empirical result: larger and better language models align better with vision models
• How Do We Ensure Alignment Of Values Betweenquestion0.769
• What is the appropriate metric for measuring representational alignment, given active debate on merits and deficiencies of all proposed measures?question0.768
  Open methodological question acknowledged as limitation
• Model Misalignmentconcept0.763
  The phenomenon of model internals deviating from desired behavior; MAS is demonstrated to detect this via comparison of toxic vs nontoxic LLMs.