Search papers, labs, and topics across Lattice.
1
0
3
LLMs struggle with raw visual tool outputs, but a simple trick鈥攖ranslating pixels into language-native summaries鈥攗nlocks massive gains in visual reasoning, outperforming even fine-tuned agents.
