Search papers, labs, and topics across Lattice.
1
0
3
GPT-4's mobile proactivity is so bad (7.4% success) that a fine-tuned Qwen2 model more than doubles its performance, revealing a critical gap in current MLLMs and a path to improvement.
