Yue Liu

The best LLM to answer a question isn't always the best LLM to *teach* the answer, and matching the "difficulty" of the explanation to the student's current abilities yields better learning.

Forget massive datasets – PilotTTS proves you can achieve state-of-the-art text-to-speech with a lean architecture and smart data engineering on just 200K hours of open-source data.

Revitalizing target-specific control within latent-query architectures for sequential CTR prediction yields consistent performance gains across diverse datasets and backbones, especially when combined with a simple position-aware reference.

Agentic AI coding assistants are alarmingly susceptible to prompt injection attacks via external artifacts, effectively turning them into remote shells for attackers.

LLM agents struggle to maintain performance in multi-day collaborative tasks, dropping significantly after just one environmental update, revealing a critical gap in adaptation to evolving real-world conditions.