Avatar

Stefan Heimersheim

Mechanistic Interpretability researcher

  1. Home
  2. Archives
  3. Search
  4. About
    1. Dark Mode
Featured image of post Removing LayerNorm
Research

Removing LayerNorm

You can remove GPT2s LayerNorm by fine-tuning for an hour

Aug 08, 2024
1 minute read
Featured image of post Sea Cucumber Essence
Research

Sea Cucumber Essence

An Exploration of feature visualization, and their relation to adversarial examples.

Jul 22, 2022
6 minute read
Featured image of post CNN Feature Visualization
Tutorial

CNN Feature Visualization

My project for AGI Safety Fundamentals programme (~Oct 2021), reproducing Chris Olah-style feature visualizations.

May 26, 2022
5 minute read
Featured image of post Fingerprint-based full-disk encryption on linux
Tutorial

Fingerprint-based full-disk encryption on linux

Using Secure Boot and TPM 2.0

Aug 30, 2021
8 minute read
Featured image of post Ask an Astronomer
Outreach

Ask an Astronomer

A weekly live-stream I organized with my Astronomy colleagues during COVID-19.

Dec 31, 2020
2 minute read
Featured image of post Digging into the Planck likelihood code
Tutorial

Digging into the Planck likelihood code

What's the `clik` library and how does it work?

Aug 07, 2020
4 minute read
© 2026 Stefan Heimersheim
Built with Hugo
Theme Stack designed by Jimmy