Search papers, labs, and topics across Lattice.
1
0
3
LLMs can save up to 40% of tokens in multi-turn reasoning by adaptively allocating compute based on turn difficulty, without sacrificing accuracy.
