Kaiser Sun

LLM agents can achieve high task accuracy while exhibiting profound epistemic *im*humility, failing to acknowledge uncertainty even when presented with contradictory evidence.

Machine translation benchmarks have functionally saturated, but pairing human-authored failure cases with deterministic verification rules reveals critical multimodal blind spots that automated metrics consistently miss.

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.