Pei Zhou

VLMs become practical preference oracles for robot learning, achieving state-of-the-art performance with drastically reduced inference costs and abstention rates.

Robots can mimic human actions but fail to grasp the underlying intent, with performance collapsing when faced with novel tasks that require true understanding.

LLMs, even when prompted or fine-tuned, struggle to replicate the messy reality of human conversation, raising serious questions about their utility as proxies for social interaction.