dtch1997

rl_cbf

Code accompanying "Value Functions are Control Barrier Functions: Verification of Safe Policies using Control Theory"

sae-probe

Investigating the feasibility of using SAE features as a basis for sparse reconstructions of linear probes

feature_composition

Experiments on feature composition in toy models and SAEs

repepo

Codebase for comparing Representation Engineering vs baselines on a variety of tasks

feature-lens

Visualizing SAE features in terms of their upstream and downstream features

steering-bench

Evaluation suite for steering vectors

advprompter

auto-circuit

A library for efficient patching and automatic circuit discovery.

belief-state-superposition

A repository for training transformers with belief states

circuit-finder

diff-interp

dtch1997.github.io

eindex

My interpretation of what einops indexing would look like (created to work on during my SERI MATS project).

factor-world

Controllable visual factors of variation for robot learning in Metaworld. Implemented in Gymnasium and pip-installable

feature-circuits

Gymnasium-Robotics

A collection of robotics simulation environments for reinforcement learning

hacking

jam

Jam - JAX models

protein-model-steering

sae-attrib-lens

sae-eap

Edge attribution patching with SAEs

sae-experiments

SAELens

Training Sparse Autoencoders on Language Models

smol-sae

stock-images

A collection of stock images for doing vision interp

SycophancySteering

Modulating sycophancy in llama-2 via activation steering

token-trace

token-trace-demo

transcoder_circuits

transcoders-slim

A minimal implementation of transcoders