Toy Models of Superposition

Can we classify further transitions in toy models?

Type: Applied

Difficulty: Easy

Status: Unstarted

This might be the easiest place to quickly get interesting new results. Try variations of the Chen et al. set-up (e.g., 3 hidden dimensions, moving away from the high-sparsity regime, varying importance, etc.). See how this affects the development of these toy models.

Where to begin:

Quantifying degeneracy (Lau et al. 2023) ,
Chen et al. 2023 ,
Notebook (start here)

If you have decided to start working on this, please let us know in the Discord. We'll update this listing so that other people who are interested in this project can find you.