From Influence Functions to Statistical Physics (draft) →

Guide for Sampling Hyperparameter Selection (draft) →

Scaling Spectroscopy Part II (draft) →

Systems, Configurations, and What We Can Observe (draft) →

Scaling Spectroscopy (draft) →

Patterning Toy Models of Superposition (draft) →

Patterning: The Dual of Interpretability

Wang, Murfet

Towards Spectroscopy: Susceptibility Clusters in Language Models

Gordon et al.

Stagewise Reinforcement Learning and the Geometry of the Regret Landscape

Elliott et al.

Focus Period: Mathematical Science of AI Safety

University of Sydney, Australia

Influence Dynamics and Stagewise Data Attribution

Lee et al.

Compressibility Measures Complexity: Minimum Description Length Meets Singular Learning Theory

Urdshals et al.

The Loss Kernel: A Geometric Probe for Deep Learning Interpretability

Adam et al.

Bayesian Influence Functions for Hessian-Free Data Attribution

Kreer et al.

ODYSSEY 2025

Berkeley, California

Embryology of a Language Model

Wang et al.

From Global to Local: A Scalable Benchmark for Local Posterior Sampling

Hitchcock, Hoogland

The Learning Process™

Modes of Sequence Models and Learning Coefficients

Chen, Murfet

Structural Inference: Interpreting Small Language Models with Susceptibilities

Baker et al.

Programs as Singularities

Murfet, Troiani

You Are What You Eat — AI Alignment Requires Understanding How Data Shapes Structure and Generalisation

Lehalleur et al.

Structure Development in List-Sorting Transformers

Urdshals, Urdshals

Dynamics of Transient Structure in In-Context Linear Regression Transformers

Carroll et al.

Differentiation and Specialization of Attention Heads via the Refined Local Learning Coefficient

Wang et al.

Loss Landscape Degeneracy and Stagewise Development of Transformers

Hoogland et al.

The Local Learning Coefficient: A Singularity-Aware Complexity Measure

Lau et al.