Transformers converge to invariant algorithmic cores
cs.LG, cs.AI
Submitted: 2026-02-26
Updated: 2026-11-09
Code: https://github.com/joshseth/cores
Terminology
Sources
- Git Re-Basin: Merging Models modulo Permutation Symmetries
- Modern Koopman Theory for Dynamical Systems
- Using Degeneracy in the Loss Landscape for Mechanistic Interpretability
- Sparse Autoencoders Find Highly Interpretable Features in Language Models
- Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models
- The Llama 3 Herd of Models
- Language Models Represent Space and Time
- Universal Neurons in GPT2 Language Models
- The Platonic Representation Hypothesis
- Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task
- Omnigrok: Grokking Beyond Algorithmic Data
- Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patching
- Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?
- Progress measures for grokking via mechanistic interpretability
- The Linear Representation Hypothesis and the Geometry of Large Language Models
- Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets
- Gemma 2: Improving Open Language Models at a Practical Size
- Open Problems in Mechanistic Interpretability
- Explaining grokking through circuit efficiency
- Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small
Related papers
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis
- (Mis)generalization of Helpful-Only Fine-tuning