News
A collection of posts written by various people associated with Developmental Interpretability (since before the agenda was conceived).
Event
The Australian AI Safety Forum 2024
11/7/2024
Paper
Differentiation and Specialization of Attention Heads via the Refined Local Learning Coefficient
10/4/2024
Blog
Singular learning theory: exercises
8/30/2024
Event
ILIAD 2024
8/28/2024
Blog
So you want to work on technical AI safety
8/24/2024
Paper
Loss landscape geometry reveals stagewise development of transformers
6/16/2024
Blog
Stagewise Development in Neural Networks
3/20/2024
Blog
Simple versus Short: Higher-order degeneracy and error-correction
3/11/2024
Blog
Timaeus's First Four Months
2/28/2024
Paper
Estimating the Local Learning Coefficient at Scale
2/6/2024
Paper
The Developmental Landscape of In-Context Learning
2/4/2024
Blog
Generalization, from thermodynamics to statistical physics
11/30/2023
Blog
Learning coefficient estimation: the details
11/15/2023
Event
The 2023 Oxford Conference
11/5/2023
Blog
Announcing Timaeus
10/22/2023
Blog
You're Measuring Model Complexity Wrong
10/11/2023
Event
The 2023 Melbourne Hackathon
10/7/2023
Event
The 2023 Amsterdam Retreat
9/18/2023
Paper
Quantifying Degeneracy in Singular Models via the Learning Coefficient
8/23/2023
Blog
DSLT 4. Phase Transitions in Neural Networks
6/24/2023
Blog
DSLT 3. Neural Networks are Singular
6/20/2023
Event
The 2023 Berkeley Conference
6/19/2023
Event
The Primer
6/19/2023
Blog
DSLT 2. Why Neural Networks obey Occam's Razor
6/18/2023
Blog
DSLT 1. The RLCT Measures the Effective Dimension of Neural Networks
6/16/2023
Blog
DSLT 0. Distilling Singular Learning Theory
6/15/2023
Blog
Approximation is expensive, but the lunch is cheap
4/19/2023
Blog
Empirical risk minimization is fundamentally confused
3/22/2023
Blog
The shallow reality of 'deep learning theory'
2/22/2023
Blog
Gradient surfing: the hidden role of regularization
2/6/2023
Blog
Interview Daniel Murfet on Universal Phenomena in Learning Machines
2/6/2023
Blog
Spooky action at a distance in the loss landscape
1/28/2023
Blog
Neural networks generalize because of this one weird trick
1/18/2023