Hao Xu

MLLMs struggle to grasp the nuances of ancient Chinese script evolution, but a glyph-driven fine-tuning approach unlocks surprisingly strong performance even in smaller models.

Turns out, MLLMs struggle with manufacturing tasks not because they can't "see," but because they lack the domain-specific knowledge to understand what they're looking at.

By explicitly modeling the latent human evaluation process, VRM offers a more robust reward model, sidestepping the pitfalls of spurious correlations that plague traditional methods.

LLMs implicitly know if their reasoning steps are correct *during* generation, according to a new step-level interpretability method.