the one thing I do have more interest in is mechanistic interpretability, because it returns meaning back to neural networks - this node represents edges, this represents a world state, and so on

Posters: pachabelcanon. pachabelcanon

2 likes recorded by the source corpus.