Search papers, labs, and topics across Lattice.
2
1
2
4
It is suggested that safety does not automatically scale with reasoning capability, especially under long-context inference, as context length increases and safety alignment consistently degrades.
Pruning redundant reasoning in teacher traces can boost recommendation model performance while streamlining output length.
