Search papers, labs, and topics across Lattice.
1
0
0
Large reasoning models don't need longer chains of thought to perform better—projecting drifting hidden states back into an efficient latent cluster cuts token generation by up to 53% while boosting task accuracy by up to 12.1%.
