Search papers, labs, and topics across Lattice.
3
0
3
5
Steering vectors may not be the universal solution for preference-aligned generation, as their effectiveness significantly drops with trait complexity and task transfer.
Localized diffusion repair can enhance summary accuracy without the overhead of complete rewrites, achieving repairs in under half a second.
LLMs don't just summarize text; they subtly rewrite narratives through biased lenses, potentially distorting the very stories we're trying to understand.
