Timothy Kostolansky

Hi, I’m Tim.

I’m a researcher interested in understanding how machine learning models learn and process information.

Currently, I’m working on language model interpretability in the Algorithmic Alignment Group; check out my Master’s thesis on some of the work! I’m also working on a mechanistic interpretability project with Jake Mendel and SPAR.

Projects/Research

A mix of the projects and research that I’ve done. Feel free to reach out to chat about any of it!

• My Master’s Thesis: Inverse Constitutional AI
• The Effect of Activation Functions On Superposition in Toy Models (with Vedang Lad)
• Iterative Interactive Inverse Constitutional AI (I^3CAI) (with Julian Manyika, code and figures)
• RL-Augmented Action Spaces in MsPacman (for 6.8200, with Julian Yocum, code to play)

Info

• This is my resume.
• I’m a member of the Algorithmic Alignment Group.

“Social” Media

• twitter
• gh
• in