Mengnan Du

DynaCF reveals that dynamically adjusting sample weights based on shortcut sensitivity can drastically improve the robustness of reward models against superficial cues.

Unlock the Rosetta Stone for neural networks: UAV lets one model explain the inner workings of *any* other, regardless of architecture or size.

LLMs respond to increasingly difficult out-of-distribution inputs by activating sparser representations in their last hidden states, revealing a quantifiable relationship between task difficulty and neural activity.

Forget complex fine-tuning: FinAnchor unlocks surprisingly robust financial predictions by simply aligning embeddings from multiple off-the-shelf LLMs.