Hongbo Xu

Language can effectively guide pixel-level anomaly detection without compromising visual fidelity, leading to unprecedented performance in industrial applications.

Sparse evidence can lead to more effective misinformation detection, with SIEVE outperforming traditional methods by focusing on critical clues rather than exhaustive analysis.

Unifying multimodal AI architectures doesn't just boost performance; it also dramatically degrades safety, especially in open-source models.