Julian Stastny

AlignmentControl
Written By Tobias Häberli

Projects

Research Direction: Studying Scheming and Alignment

  • Studying abstract analogies of scheming such as these two proposed projects.

  • Training AIs to be more goal-directed and coherent (and studying whether this can lead to scheming).

  • Training AIs on various model specs and studying effects on the model's personality.

What I'm looking for in a Mentee

  • Strong empirical research skills.

  • Ability and drive to take conceptual ownership of their project.

  • Enjoys the early exploratory part of research (where there aren't any numbers to drive up yet).

Bio

At Redwood, I manage a bunch of empirical and conceptual research projects. Previously I worked at the Center on Long-term Risk. My academic background is in computer science and machine learning.

Tobias Häberli
Next

Oscar Delaney