paper

Contrastive-Difference CKA Reveals Concept-Specific Structural Alignment Across Language Model Architectures

paperactiveprovisionalcontrastive-difference-cka-reveals-concept-specific-structural-alignment-across-language-model-architectures-1d724a70·1 events·first seen 25h ago

Aliases: Contrastive-Difference CKA Reveals Concept-Specific Structural Alignment Across Language Model Architectures

Co-occurring entities

CKA_Delta Gemma

More like this (12)

contrastive semantic alignment Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models The Neutral Mask: How RLHF Provides Shallow Alignment while Leaving Partisan Structure Intact in a Large Language Model Latent Context Language Models Transformer Language Models Contrastive Pre-training Acoustic Cue Alignment in Audio Language Models for Speech Emotion Recognition The Shibboleth Effect: Auditing the Cross-Lingual Distributional Skew of Large Language Models Reasoning Language Models Multi-Faceted Interactivity Alignment in Full-Duplex Speech Models Adaptive Multi-Resolution Procedural Knowledge Compression for Large Language Models Contrastive Search

Recent events (1)

5arXiv · cs.CL·25h ago·source ↗

Contrastive-Difference CKA reveals concept-specific structural alignment across LLM architectures

Researchers introduce CKA_Delta (contrastive-difference CKA), a training-free diagnostic that isolates concept-specific representational convergence from generic similarity across LLM architectures. The method reveals a geometric-functional universality dissociation: moderate geometric alignment coexists with near-perfect functional transfer across six concept domains and multiple architectural families. CKA_Delta also functions as an architectural outlier detector, flagging Gemma as a notable outlier (d=1.08, AUC=0.79). The work provides a practical tool for cross-architecture concept monitoring without requiring model training.

Evaluation and Benchmarking AI Safety Research CKA_Delta Gemma Contrastive-Difference CKA Reveals Concept-Specific Structural Alignment Across Language Model Architectures