neat idea! Would be really cool to have a tool for quickly identifying structural similarities between neural networks
nice! It’s great to see steps towards interpretability of multi modal models
cool and ambitious idea. I would be a bit hesitant with impossibility claims like “if it’s that many significant eigen valued then it’s not interpretable” because there’s always a chance of new techniques changing the perspective
cool idea to approach it from the side of general linguistic patterns. and very ambitious, I like it
