Search papers, labs, and topics across Lattice.
1
0
2
3
LLMs can be steered to better logical reasoning by exploiting a shared, low-dimensional "logical subspace" learned by aligning natural language and symbolic reasoning views.