Hacker News Reader

Reminds me of LLMs, where the weights are the 'seed' that solves standard benchmark suite.

The author treats the seed as a hyperparameter and searches for the one that performs best for training a CV model.