Search papers, labs, and topics across Lattice.
Department of Computer Science, Data Science and AI Institute, Johns Hopkins University, Johns Hopkins University
2
0
6
6
LLMs learn skills in a surprisingly consistent order during pretraining, revealing a hidden curriculum that's predictable across models and readable from their internal representations.
MLLMs struggle with visually rendered text not because they can't reason, but because they can't *read* it, and a simple self-distillation fix closes the gap.
