Search papers, labs, and topics across Lattice.
Peking University
1
0
3
5
Forget reward engineering: selectively updating the most "active" weights in LLMs can unlock significant gains in long-context reinforcement learning.
