Search papers, labs, and topics across Lattice.
β Xiaohongshu Inc, Fudan University, Tongji University, δ University of California
2
0
6
Frontier models can't build playable games in one shot, but a closed-loop system using GUI agents to playtest and refine code achieves a 66.8% success rate, proving that game generation needs to be a conversation, not a translation.
Forget CLIP: initializing vision encoders from text-only LLMs unlocks surprising gains in visual fidelity and data efficiency for VLMs, rivaling or surpassing larger, contrastively-pretrained models.
