A Theory of Deep Learning | Hacker News
TL;DR AI
2 min readKey summary
A Hacker News commenter criticized a proposed unified theory of deep learning as too broad and circular.
They argued it mostly re-labels memorized content as signal vs. noise, without explaining why SGD learns that separation.
The critique says the field needs stronger evidence, testable predictions, and clearer links to generalization before grand unification claims.
The debate touches on issues like benign overfitting, double descent, grokking, and implicit bias.
Same topic
