Hi, I’m Stefan! I try to understand neural networks & LLMs by analysing their internals (“mechanistic interpretability”). I’ve worked at various AI safety organizations including Apollo Research where I develop new mechanistic interpretability tools, and FAR.AI where I explored using interpretability to improve safety.

Contact

The best way to get in touch with me is via email (firstname.lastname@gmail.com), messaging me on the Open Source Mech Interp Slack, or a DM on LessWrong.

Legal

Responsible for the content on this site is

Stefan Heimersheim
25 Holywell Row
EC2A 4XE
London

External links disclaimer: This website may link to third-party websites. I have no control over their content or privacy practices and accept no responsibility for them. Visiting those sites is at your own risk.