Intrinsic Dimensions: How Learning in Large Models Is Driven by a Few Parameters | Hacker News Reader