I-Hung Hsu

Composing aesthetic preference with explicit rubric rewards unlocks substantial gains in text-to-image generation, outperforming existing open-source models by a significant margin.

Gemma 4's unified architecture and reasoning mode enable it to outperform larger models in human-rated tasks while maintaining high efficiency.

Existing benchmarks miss the mark on faithfulness, but a new dependency-aware checklist reveals the true performance gaps in T2I models.

Forget hand-annotated data: Magnet distills multi-turn tool-use skills into LLMs by automatically generating training trajectories that outperform even Gemini 1.5 Pro.