5 ms·
Toy Models of Superposition (2022)
- dustingetz 3y agoguys. the mobile layout. people consume hobby content on mobile
- naillo 3y agoAnthropic is awesome
- jiggawatts 3y agoThis is one of the best presentations of such a complex topic, or any topic for that matter that I've seen in a long time! To the authors if they happen to find themselves here, I say: bravo!
- DigitalNoumena 3y agoI am curious why isn't the network always in an "interference" state, but sometimes it collapses into a "restricted" ONB? > the neural networks we observe in practice are in some sense noisily simulating larger, highly sparse networks This seems somewhat related to a point made by Ilya Sutskever here [1]: NNs can be though of an approximation to the Kolmogorov compressor. Speculating, one could say any network is a projection of the ideal compressor (which argualy perfectly represents all n-features in an n-dimensional ONB) into a lower dimentional space, hence the interference. But why is there not always such an interference? [1] https://www.youtube.com/watch?v=AKMuA_TVz3A https://www.youtube.com/watch?v=AKMuA_TVz3A