Zixin Ding

Agents can exhibit significant performance gains from retained experience, but the pathways to these improvements are often unclear and model-dependent.

Reinforcement learning can boost signal efficiency in high-throughput scientific facilities, achieving a 56% improvement in real-time event filtering at the Large Hadron Collider.

Even GPT-5.1 struggles to distinguish AI-generated academic images from real ones, achieving only 48.8% accuracy, revealing a significant gap between generative and forensic AI capabilities.

Data valuation for LLMs doesn't need backpropagation: a single forward pass is enough to match gradient-based methods, unlocking massive speedups.