Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis
cs.CV, cs.LG, cs.NE
Submitted: 2025-05-16
Updated: 2026-10-23
Code: https://github.com/akarshkumar0101/fer
Terminology
Sources
- GPT-4 Technical Report
- Mechanistic Interpretability for AI Safety -- A Review
- Understanding disentangling in $\beta$-VAE
- Training Verifiers to Solve Math Word Problems
- Toy Models of Superposition
- General Intelligence Requires Reward-based Pretraining
- Training Compute-Optimal Large Language Models
- Open-Endedness is Essential for Artificial Superhuman Intelligence
- The Platonic Representation Hypothesis
- A Theory of Universal Artificial Intelligence based on Algorithmic Complexity
- On the "steerability" of generative adversarial networks
- Mixtral of Experts
- Scaling Laws for Neural Language Models
- Evolution and The Knightian Blindspot of Machine Learning
- DeepSeek-V3 Technical Report
- Omnigrok: Grokking Beyond Algorithmic Data
- Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solve
- GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models
- Illuminating search spaces by mapping elites
- Progress measures for grokking via mechanistic interpretability
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs