Search papers, labs, and topics across Lattice.
1
0
3
5
Forget test-time training: this work bakes optimal control directly into LLMs, yielding up to 27.8% gains in mathematical reasoning.
