Search papers, labs, and topics across Lattice.
1
0
3
LLMs encode cognitive complexity from basic recall to abstract synthesis in linearly separable subspaces, suggesting a surprisingly structured internal representation of thought.
