Search papers, labs, and topics across Lattice.
1
0
2
Alignment training doesn't erase political bias鈥攊t merely conceals a measurable latent direction that leads models to hardcode transient partisan consensus as timeless objective fact.
