Search papers, labs, and topics across Lattice.
Shanghai Jiao Tong University
2
0
5
Current audio editing models are failing spectacularly, with an Exact Match Rate below 5% in complex tasks, exposing a critical need for improvement.
Even state-of-the-art T2I models falter in generating scientifically accurate illustrations, revealing critical gaps in text-rendering and reasoning capabilities.
