Search papers, labs, and topics across Lattice.
Huazhong University of Science and Technology
2
0
5
Proprietary MLLMs generate visually appealing code from Figma designs, but fall short on crucial aspects like responsive layouts and maintainability, revealing a surprising gap in their practical utility.
Ground-truth labels don't improve LLM reasoning, but building a consensus-based knowledge graph of reasoning steps does, boosting accuracy by 10%+.
