Chris Thomas

dLVLMs not only reverse the yes-bias of AR models but also collapse in accuracy for underrepresented groups, revealing critical reliability gaps in diffusion-based approaches.

Introspective signals from LVLMs can provide stronger factuality guarantees than traditional external verification methods, leading to fewer incorrect outputs.

Grounded event discovery in long videos is a daunting challenge, with event trigger detection rates plummeting below 8%.