Mechanistic Interpretability

Letting the Neural Code Speak: Automated Characterization of Monkey Visual Neurons Through Human Language

We develop digital twins of V1 and V4 neurons and use generative models to translate neural activity patterns into semantic descriptions, driving 96.1% of V4 neurons to extreme activation levels with synthesized images based on linguistic descriptions.

Mechanistic Interpretability for Progress Towards Quantitative AI Safety

MIT Master’s thesis studying mechanistic interpretability as a path toward quantitative AI safety.