Search papers, labs, and topics across Lattice.
1
0
2
LLMs can significantly improve their performance on complex tasks like math and coding *without any external rewards*, simply by iteratively comparing and refining their own outputs.
