Concepts Whisper: Spectral Anti-Concentration and the Dual Geometry of Transformer Representations
cs.LG, cs.AI
Submitted: 2026-05-02
Updated: 2026-08-29
Code: https://github.com/pratyush-acharya/spectral_anti_concentration
Terminology
Sources
- Refusal in Language Models Is Mediated by a Single Direction
- Language Models Represent Space and Time
- Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders
- Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders
- Language-Specific Latent Process Hinders Cross-Lingual Performance
- Polysemanticity and Capacity in Neural Networks
- The Truth is in There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction
- Small Singular Values Matter: A Random Matrix Analysis of Transformer Models
- Whitening Sentence Representations for Better Semantics and Faster Retrieval
- Linguistic Collapse: Neural Collapse in (Large) Language Models
- Representation Engineering: A Top-Down Approach to AI Transparency
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks