Skip to content

Doing The Thing

Dyck Transformer Probe

Initializing search

Doing The Thing

Doing The Thing: From Mechanical Turk to Mech Interp
Dlog
Dlog
Notes
Notes
- Reference Links
- Running Todo List
- Concepts
  Concepts
  - Einsum Guide for Attention Pattern
  - Stats and Probability Concepts
- Papers
  Papers
  - Information theory
    Information theory
    
    Revisting the Uniform Information Density Hypothesis
  - Interpretability
    Interpretability
    
    Towards Automated Circuit Discovery for Mechanistic Interpretability
    
    Automated Interpretability Agenda
    
    Mixture of Linear Transforms (MOLTs)
    
    Interpreting Context Look Ups
    
    Detecting and reducing scheming in AI models
- Projects
  Projects
- Reflections
  Reflections

Dyck Transformer Probe

Work in progress ntoes for probing models for syntactic dependency

Linear Subspace

March 31, 2026

Detecting and reducing scheming in AI models

Multilingual Semantics Probe

kgng [at] usc [dot] edu

Made with Material for MkDocs