Li Miao

I lead Multimodal Content Understanding for YouTube's social ecosystems (Comments, Posts, Live Chat). My systems analyze Text, Visuals, and Context signals to decode semantic meaning at massive scale — models that look beyond keywords to understand the substance of community interactions and power core safety decisions.

I apply this deep understanding to solve AI Responsibility challenges. By accurately detecting nuance in complex social data, my models empower Creator Agency — giving creators precise, automated tools to filter harmful noise and amplify high-quality engagement, effectively acting as personalized "moderation agents" for their communities.

To me, these systems are precursors to Autonomous Agents: they perceive a multimodal environment, interpret intent, and take actions without human intervention. I am applying the lessons from stabilizing these large-scale systems — specifically around adversarial robustness and alignment — to solve the broader challenge of Agent Reliability.