Search papers, labs, and topics across Lattice.
1
0
2
LLMs can compress their reasoning by over 50% with minimal accuracy loss by selectively applying latent compression only to redundant steps.
