Jeff Schneider

We scale the influence-function-based data valuation method to recent LLMs and their massive training datasets.

Encouraging model in model-based reinforcement learning to converge to flatter minima in the loss landscape will result in better downstream policies