VibesWire

Ctrl + K
Log In
Pruning Training Data by Loss Alone Makes a 110M-Parameter Model Memorize Facts as Well as a 1.3B Model — With an Information-Theoretic Proof for Why | VibesWire