Interpretability

The Problem

Neural networks are largely black boxes: we see inputs and outputs, but not the internal logic connecting them. If we don't know why a model made a decision, we can't be confident it will behave safely in novel situations, and existing methods struggle to scale as models grow to billions of parameters.

What We're Working On

• Mechanistic interpretability — Finding circuits in small-to-medium models that generalize to larger systems.
• Representation & self-supervised learning — What learned representations encode and how they transfer across tasks.
• Activation steering — Using internal representations to monitor and control model behavior at inference time.